BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1032
DTSTAMP:20230914T125947Z
SUMMARY:Online Parameter Estimation for Human Driver behavior Prediction fr
om Driving Demonstration Data
DESCRIPTION:Speaker: Raunak P. Bhattacharyya (Stanford University\, USA)\n\
nAbstract: \nAbstract: Driver models are invaluable for safety validation
in simulation. However\, driver modeling is characterized by a high degree
of uncertainty. While rule-based driver models have the advantage of bein
g interpretable and collision-free\, black-box models are more expressive
than rule-based models and capture more nuanced behavior. Unfortunately\,
such black-box models lack interpretablity and fail to incorporate the saf
ety guarantees of the rule-based models. While most approaches in driver-m
odeling literature select model parameters offline\, online estimation has
the advantage of being able to capture the behavior of individual drivers
. In this paper\, we show that online parameter estimation applied to the
Intelligent Driver Model captures nuanced individual driving behavior whil
e providing collision free trajectories. We benchmark performance against
rule-based and black-box driver models on two real world driving data sets
. We evaluate the closeness of our driver model to ground truth data demon
stration and also assess the safety of the resulting emergent driving beha
vior.\n
URL:https://www.tcs.tifr.res.in/web/events/1032
DTSTART;TZID=Asia/Kolkata:20200102T143000
DTEND;TZID=Asia/Kolkata:20200102T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1033
DTSTAMP:20230914T125947Z
SUMMARY:Workshop on Learning Theory 2
DESCRIPTION:Speaker: \n\nAbstract: \nThe field of Learning Theory has witne
ssed an explosive growth over the past decade. This growth has been made p
ossible by exciting developments in the fields of applied mathematics\, st
atistics and computer science and fueled by the need for processing modern
large data sets.\nThis two-day workshop seeks to explore some aspects of
statistical learning theory. It will include five mini-courses and featu
re a couple of talks.\nParticipation is by invitation only. Please conta
ct learningtheory@tifr.res.in for further information.\n
URL:https://www.tcs.tifr.res.in/web/events/1033
DTSTART;TZID=Asia/Kolkata:20200103T090000
DTEND;TZID=Asia/Kolkata:20200104T173000
LOCATION:AG-66 (Lecture Theatre)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1034
DTSTAMP:20230914T125947Z
SUMMARY:Bombay Information Theory Seminar 2020
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1034
DTSTART;TZID=Asia/Kolkata:20200106T090000
DTEND;TZID=Asia/Kolkata:20200109T170000
LOCATION:TIFR and IIT Bombay
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1035
DTSTAMP:20230914T125948Z
SUMMARY:Testing Noisy Linear Equations for Sparsity
DESCRIPTION:Speaker: Anindya De (University of Pennsylvania\nUnited States)
\n\nAbstract: \nAbstract: Consider the following basic problem in sparse l
inear regression -- an algorithm gets labeled samples of the form (x\, +
\\eps) where w is an unknown n-imensional vector\, x is drawn from a bac
kground distribution D and \\eps is some independent noise. Given the pro
mise that w is k-sparse\, the breakthrough work of Candes\, Rhomberg and
Tao (2005) shows that w can be recovered with samples and time which scal
es as O(k log n). This should be contrasted with general linear regressio
n where O(n) samples are information theoretically necessary.\nIn this
talk\, we look at this question from the vantage point of property testin
g and study the decision variant of the following question -- namely\, wh
at is the complexity of deciding if the unknown vector w is k-sparse (or
at least say 0.01 far from k-sparse in \\ell_2 distance). We show that th
e decision version of the problem can be solved with samples which are in
dependent of n as long as the background distribution D is i.i.d. and the
components are not Gaussian. We further show that weakening any of the c
onditions in this result necessarily makes the complexity scale as log n
(thus showing our results are tight).\nJoint work with Xue Chen (Northwes
tern) and Rocco Servedio (Columbia).\n
URL:https://www.tcs.tifr.res.in/web/events/1035
DTSTART;TZID=Asia/Kolkata:20200106T103000
DTEND;TZID=Asia/Kolkata:20200106T113000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1036
DTSTAMP:20230914T125948Z
SUMMARY:Learning and Characterization of Interventional Equivalence class o
f Causal Graphs with Latents
DESCRIPTION:Speaker: Karthikeyan Shanmugam (IBM Research AI Group\, USA)\n\
nAbstract: \nAbstract: Directed Causal Graphs (DAGs) capture causal relati
onships amongst a set of variables and they specify how interventional dis
tributions relate to observational ones. Unobserved latent variables are
represented in DAGs with edges having double arrows. Celebrated``do-calcu
lus” introduced by Pearl relates invariances in interventional distribut
ions to a specific causal graph with latents embodying expert knowledge. W
e consider the reverse problem of learning the equivalence class of causal
graphs that could imply the observed invariances of the do-calculus. Give
n observational and interventional data obtained under soft interventions
with known targets\, we provide a complete characterization of the equival
ence class of Causal DAGs. We also provide a sound learning algorithm to l
earn the equivalence class under additional faithfulness assumptions.\nBio
:\nKarthikeyan Shanmugam is currently a Research Staff Member with the IBM
Research AI group in NY. Previously\, he was a Herman Goldstine Postdocto
ral Fellow in the Math Sciences Division at IBM Research\, NY. He obtained
his Ph.D. in Electrical and Computer Engineering from UT Austin in 2016\,
MS degree in Electrical Engineering from USC in 2012\, B.Tech and M.Tech
degrees in Electrical Engineering from IIT Madras in 2010. His research in
terests broadly lie in Graph algorithms\, Machine learning\, Optimization\
, Coding Theory and Information Theory. In machine learning\, his research
focus is primarily on causal inference\, online learning and explainable
ML. He has also worked on problems relating to information flow\, storage
and caching over networks.\n
URL:https://www.tcs.tifr.res.in/web/events/1036
DTSTART;TZID=Asia/Kolkata:20200110T143000
DTEND;TZID=Asia/Kolkata:20200110T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1037
DTSTAMP:20230914T125948Z
SUMMARY:PIDForest: Anomaly detection via Partial Identification
DESCRIPTION:Speaker: Parikshit Gopalan (VMware Research\nPalo Alto\, CA\, U
.S.)\n\nAbstract: \nAbstract: Anomaly detection is a ubiquitous problem in
machine learning. Here one is given a large population of points\, we may
not have much knowledge about their structure a priori. The goal is to fi
gure out what is "typical" of the datatset\, and what points are atypical
or anomalous. This talk will explore a framework called Partial Identifi
cation for identifying anomalies in a large datatset.\n\nWe propose a defi
nition for “anomalousness” that captures the intuition that anomalies
are easy to distinguish from the overwhelming majority of points by relati
vely few attribute values: we call this partial identification. Our notio
n is inspired by the notion of Partial IDs that were studied by Yehudayoff
and Wigderson in the context of population recovery. Formalizing this int
uition\, we propose a geometric anomaly measure for a point that we call P
IDScore\, which measures for the minimum density of data points over all s
ubcubes containing the point. We present PIDForest: a random forest based
algorithm that finds anomalies based on this definition and show that it
performs favorably in comparison to several popular anomaly detection meth
ods\, across a broad range of benchmarks.\n\nBased on joint work with Udi
Wieder (VMware) and Vatsal Sharan (Stanford) that appeared in NeurIPS 2019
.\n
URL:https://www.tcs.tifr.res.in/web/events/1037
DTSTART;TZID=Asia/Kolkata:20200113T143000
DTEND;TZID=Asia/Kolkata:20200113T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1038
DTSTAMP:20230914T125948Z
SUMMARY:Privacy-Preserving Localization and Recognition of Human Activitie
s
DESCRIPTION:Speaker: Prof. Janusz Konrad (Boston University\, USA)\n\nAbstr
act: \nAbstract: Smart rooms\, that respond to occupant behavior\, will l
ikely become a common occurrence in our lifetimes. With advanced sensors\
, processors and algorithms\, such rooms are expected to save energy and
provide productivity as well as health benefits. Indoor localization of
occupants and recognition of their activities are two key components of t
his vision. However\, traditional camera-based systems may not be accepta
ble in privacy-sensitive scenarios since high-resolution images may revea
l room details and occupant identity to eavesdroppers. I will first revie
w traditional approaches to protecting occupant’s visual privacy\, incl
uding reversible methods (e.g.\, data scrambling) and irreversible method
s (e.g.\, optical and digital obfuscation of visual data). Then\, I will
describe computational solutions via resolution reduction recently develo
ped at Boston University. In one approach\, a small network of ceiling-mo
unted\, single-pixel RGB sensors collects visual data from which occupant
’s location is estimated by means of advanced algorithms. In another ap
proach\, extremely low resolution videos (e.g.\, 16 x 12 pixels) are used
to recognize occupant’s activities. Finally\, I will discuss visual pr
ivacy protection by replacing the identity information in a person’s im
age with another identity by means of variational generative-adversarial
networks (VGANs).\nBio: Janusz Konrad received Master’s degree from Tec
hnical University of Szczecin\, Poland in 1980 and PhD degree from McGill
University\, Montréal\, Canada in 1984. He joined INRS-Télécommunicat
ions\, Montréal as a post-doctoral fellow and\, since 1992\, as a facult
y member. Since 2000\, he has been on faculty at Boston University. He is
an IEEE Fellow and a recipient of several IEEE and EURASIP Best Paper aw
ards. He has been actively engaged in the IEEE Signal Processing Society
as a member of various boards and technical committees\, as well as an or
ganizer of conferences. He has also been on editorial boards of various E
URASIP journals. His research interests include video processing and comp
uter vision\, stereoscopic and 3-D imaging and displays\, visual sensor
networks\, human-computer interfaces\, and cybersecurity.\n
URL:https://www.tcs.tifr.res.in/web/events/1038
DTSTART;TZID=Asia/Kolkata:20200113T160000
DTEND;TZID=Asia/Kolkata:20200113T170000
LOCATION:D-Block Conference Room (D-406)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1039
DTSTAMP:20230914T125948Z
SUMMARY:Solidity+: A Language for Robust Programming of Smart Contracts
DESCRIPTION:Speaker: R.K. Shyamasundar (IIT Bombay\, Mumbai)\n\nAbstract: \
n"2nd UNESCO World Logic Day Celebration Lecture"\nAbstract: Smart Contra
cts handle and transfer assets of considerable value. Thus\, it is crucial
that their implementation be secure against attacks which aim at stealin
g or tampering the assets. In the recent past\, there have been several a
ttacks that have exploited existing vulnerabilities in smart contracts. O
nce a smart contract is up and running\, changing it is very complicated a
nd nearly infeasible. One of the reasons is that when a contract is create
d\, it is immutable\; once deployed on the Blockchain it stays there fore
ver. If we find a defect in a deployed smart contract\, a new version o
f the contract has to be created and deployed. When we deploy a new versio
n of an existing contract\, data stored in the previous contract does not
get transferred automatically to the newly refined contract. We have t
o manually initialize the new contract with the past data which makes it
very cumbersome. Similarly\, neither updating a contract nor rolling back
an update is possible\; this greatly increases the complexity of implemen
tation and places a huge responsibility while being deployed initially on
the Blockchain. We describe the design and use of language Solidity+\, fo
r programming smart contracts. An important outcome of using of Solidity+
\, is that brings out an outline of a proof carrying code for the smart c
ontract for free – needless to emphasize that it is a very welcome feat
ure for smart contracts on Blockchains.\n
URL:https://www.tcs.tifr.res.in/web/events/1039
DTSTART;TZID=Asia/Kolkata:20200114T143000
DTEND;TZID=Asia/Kolkata:20200114T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1040
DTSTAMP:20230914T125948Z
SUMMARY:Formalizing Finite Set Combinatorics in Type Theory
DESCRIPTION:Speaker: Abhishek Singh\n\nAbstract: \nAbstract: Mathematical p
roofs when written in conventional ways often contain imprecise definition
s\, unstated background assumptions\, and inferential gaps in reasoning. I
n such circumstances\, it becomes difficult for a reviewer to determine wh
ether the given proof is correct or not. Even if the theorem statement t
urns out to be true\, judging it to be so could take a long time. A possib
le solution to this problem is to formalize mathematical results using Pro
of Assistants. Proof Assistants are software tools built on top of a small
and trusted kernel that provides a formal language for writing mathematic
al statements and their proofs. Hence\, formally verifying a mathematical
theory using a Proof Assistant can increase our confidence in the verified
results. However\, the task of formalizing mathematics using Proof Assist
ants presents some unique challenges\; both practical as well as theoretic
al. Theoretical challenges mostly arise because some trivially assumed axi
oms of classical mathematics may not be provable in the core logic of the
Proof Assistant. On the other hand\, the practical difficulties of formali
zation mostly arise because the machine-checkable proofs are significantly
more detailed than the corresponding paper proofs. In this talk\, we will
address both these issues while presenting formal proofs of some key resu
lts from finite set combinatorics. More precisely\, we present formalized
libraries of definitions and results on two important mathematical structu
res from combinatorics: (i) finite partially ordered sets\, and (ii) finit
e simple graphs. These libraries have been formalized in the type theory o
f Coq Proof Assistant.\n
URL:https://www.tcs.tifr.res.in/web/events/1040
DTSTART;TZID=Asia/Kolkata:20200115T113000
DTEND;TZID=Asia/Kolkata:20200115T130000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1041
DTSTAMP:20230914T125948Z
SUMMARY:A Quadratic Lower Bound for Algebraic Branching Programs
DESCRIPTION:Speaker: Prerona Chatterjee\n\nAbstract: \nAbstract: An Algeb
raic Branching Program (ABP) is a layered graph where each edge is labele
d by an affine linear form and the first and the last layer have one vert
ex each\, called the “start” and the “end” vertex respectively. T
he polynomial computed by an ABP is equal to the sum of the weights of al
l paths from the start vertex to the end vertex in the ABP\, where the we
ight of a path is equal to the product of the labels of all the edges on
it. The size of an ABP is the number of vertices in it.\nIn this talk\, we
will see a proof of the following result. Any Algebraic Branching Progra
m (ABP) computing the polynomial $\\sum_{i = 1}^n x_i^n$ has at least $\\
Omega(n^2)$ vertices.\nThis improves upon the lower bound of $\\Omega(n \\
log n)$\, which follows from the classical result of Baur and Strassen [S
tr73\, BS83]\, and extends the results in [Kum19]\, which showed a quadra
tic lower bound for homogeneous ABPs computing the same polynomial.\nThe
talk is based on work done with Mrinal Kumar\, Adrian She and Ben Lee Vol
k.\n
URL:https://www.tcs.tifr.res.in/web/events/1041
DTSTART;TZID=Asia/Kolkata:20200117T160000
DTEND;TZID=Asia/Kolkata:20200117T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1042
DTSTAMP:20230914T125948Z
SUMMARY:Deterministic Representation of Linear Matroids
DESCRIPTION:Speaker: Pranabendu Misra (Max Planck Institute for Informatics
\nSaarbrucken\, Germany.)\n\nAbstract: \nAbstract: Matroids are combinato
rial objects that generalize the notion of linear independence. They have
several applications in design and analysis of algorithms. Linear matroi
ds are a subclass of matroids that can be represented by a matrix. Recent
ly\, these matroids have found applications in Parameterized Complexity\,
including some breakthrough results. In this talk\, we will discuss the
problem of constructing a matrix representation of linear matroids\, espe
cially via deterministic algorithms.\n
URL:https://www.tcs.tifr.res.in/web/events/1042
DTSTART;TZID=Asia/Kolkata:20200121T143000
DTEND;TZID=Asia/Kolkata:20200121T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1043
DTSTAMP:20230914T125948Z
SUMMARY:Ultra Large-scale Linear Programming based on Continuum Computing
DESCRIPTION:Speaker: Dr. Narendra Karmarkar (Former Homi Bhabha Chair Prof.
\, TIFR)\n\nAbstract: \nAbstract: Ultra Large-scale Linear Programming re
fers to class of problems where number of linear inequality constraints gr
ows exponentially w.r.t. the number of variables. "Continuum Computing" is
a generalization of powerful interior point algorithms\, and involves com
puting with transcendental functions\, often in several complex variables.
\n\nThere are many potential applications of this work to several practi
cal problems-e.g. Explainable AI\, Topology optimization for 3D printing\,
Optimal mask design for 7nm lithography etc. However\, our focus in this
seminar will be on core mathematical technique and it's parallelization on
contemporary parallel machines.\n\nWe show that the "Potential function"
and it's first two derivatives\, involved in the algorithm for certain cla
ss of ultra large scale LPs\, can be computed in polynomial time in the co
ntinuum computing paradigm\, in spite of the fact that the feasible region
is given by exponential number of constarints. Actual numerical cross-sim
ulation on standard computing model and machines involves approximate nume
rical computation of inverse Laplace transform. We also apply this approac
h to establising non-satifiability of boolean formula\, which is known to
require exponentially long resolution-based proofs in the standard Turing
machine model. It is also believed that this task can not be done efficien
tly in quantum computing model. Earlier exposition of this work from FOCM
can be found in Cornell Arxive 1412.3335.pdf\, and LNCS6457.\n
URL:https://www.tcs.tifr.res.in/web/events/1043
DTSTART;TZID=Asia/Kolkata:20200121T160000
DTEND;TZID=Asia/Kolkata:20200121T170000
LOCATION:AG-66 (Lecture Theatre)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1044
DTSTAMP:20230914T125948Z
SUMMARY:Learning in Gated Neural Networks
DESCRIPTION:Speaker: Makkuva Ashok Vardhan (Department of Electrical & Comp
uter Engineering\nUniversity of Illinois at Urbana-Champaign (UIUC)\nUnite
d States)\n\nAbstract: \nAbstract: Gating is a key feature in modern neur
al networks including LSTMs\, GRUs and sparsely-gated deep neural networks
. The backbone of such gated networks is a mixture-of-experts layer\, wher
e several experts make regression decisions and gating controls how to wei
gh the decisions in an input-dependent manner. Despite having such a promi
nent role in both modern and classical machine learning\, very little is u
nderstood about parameter recovery of mixture-of-experts since gradient de
scent and EM algorithms are known to be stuck in local optima in such mode
ls. In this work\, we perform a careful analysis of the optimization lands
cape and show that with appropriately designed loss functions\, gradient d
escent can indeed learn the parameters accurately. A key idea underpinning
our results is the design of two distinct loss functions\, one for recove
ring the expert parameters and another for recovering the gating parameter
s. We demonstrate the first sample complexity results for parameter recove
ry in this model for any algorithm and demonstrate significant performance
gains over standard loss functions in numerical experiments. \n\nBio: A
shok is a 5th year graduate student in the ECE department at UIUC\, advise
d by Prof. Pramod Viswanath. He obtained his Masters in ECE (advised by Pr
of. Yihong Wu) from UIUC in 2017 and Bachelors in EE (advised by Prof. Viv
ek Borkar) with a minor in Mathematics from IIT Bombay in 2015. His curren
t research interests are theoretical and algorithmic aspects of machine le
arning and information theory. He is a recipient of Best Paper Award from
ACM MobiHoc 2019. He has won several graduate student awards and fellowshi
ps including Joan and Lalit Bahl Fellowship\, Sundaram Seshu International
Student Fellowship\, and was also a finalist for the Qualcomm Innovation
Fellowship 2018. Outside the convex hull of research activities\, he likes
to learn new languages\, watch and read about international films\, readi
ng history\, and remembering trivia. For more details\, please visit: http
://makkuva2.web.engr.illinois.edu/\n
URL:https://www.tcs.tifr.res.in/web/events/1044
DTSTART;TZID=Asia/Kolkata:20200123T143000
DTEND;TZID=Asia/Kolkata:20200123T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1045
DTSTAMP:20230914T125948Z
SUMMARY:Paper: The Capacity of Wireless Networks
DESCRIPTION:Speaker: Kumar Saurav\n\nAbstract: \nGupta\, P.\, & Kumar\, P.
R. (2000). The capacity of wireless networks.\nIEEE Transactions on inform
ation theory\, 46(2)\, 388-404.\n
URL:https://www.tcs.tifr.res.in/web/events/1045
DTSTART;TZID=Asia/Kolkata:20200124T140000
DTEND;TZID=Asia/Kolkata:20200124T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1046
DTSTAMP:20230914T125948Z
SUMMARY:Weight Distribution and List-Decoding size of Reed-Muller codes
DESCRIPTION:Speaker: Tulasi mohan Molli\n\nAbstract: \nAbstract: The proble
m of list-decoding an error-correcting code is the following:\ngiven a rec
eived word and a distance parameter find all codewords of the code that ar
e within the given distance from the received word. It is a generalization
of the more common notion of unique decoding.\nThe weight distributio
n of an error-correcting code counts\, for every given weight parameter\,
the number of codewords whose hamming weight is less than the given weight
parameter.\nThe codewords of Reed-Muller code can be thought of as truth-
tables of low degree polynomials. Kaufman\, Lovett\, and Porat in their wo
rk from 2010 made a novel connection between computer science techniques u
sed for studying low-degree polynomials and these seemingly related coding
theory questions in the case of Reed-Muller codes.\nIn this talk\, we wil
l see\n1) The above-mentioned result of Kaufman\, Lovett\, and Porat[KLP10
] and subsequent improvement of it due to Abbe\, Sphilka\, and Wigderson[A
SW15] which give upper bounds on the weight distribution of Reed-Muller co
des.\n2) A lower bound on the weight-distribution of Reed-Muller codes by
exhibiting a collection of low-degree polynomials that satisfy the weight
requirement for any given weight parameter. This is joint work with Bhanda
ri\, Harsha\, and Saptarishi and independently discovered by Sberlo and Sh
pilka[SS20].\n
URL:https://www.tcs.tifr.res.in/web/events/1046
DTSTART;TZID=Asia/Kolkata:20200124T160000
DTEND;TZID=Asia/Kolkata:20200124T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1047
DTSTAMP:20230914T125948Z
SUMMARY:Towards Optimal Secure Computation Protocols
DESCRIPTION:Speaker: Akshayaram Srinivasan (University of California\nBerke
ley\, CA\, United States)\n\nAbstract: \nAbstract: Secure computation all
ows a set of mutually distrusting parties to compute a joint function of t
heir private inputs such that the parties only learn the output of the fun
ctionality and nothing else about the inputs of the other parties. Secure
computation is one of the central primitives in cryptography that encompas
ses several cryptographic abstractions and has many practical applications
. The seminal results from the 1980s showed that every efficiently computa
ble functionality can also be computed securely. However\, these protocols
were prohibitively inefficient and could only be considered as feasibilit
y results. One of the central problems in cryptography is to construct sec
ure computation protocols that are optimal in all efficiency parameters. I
n this talk\, I will give an overview of my recent works that make progres
s towards constructing such optimal secure computation protocols.\nBio: Ak
shayaram Srinivasan is a final-year Ph.D. student in the theory group at U
C Berkeley. He received his B.Tech in Computer Science and Engineering fro
m IIT-Madras in 2015. He is broadly interested in theoretical computer sci
ence and in particular\, in the theory and applications of cryptography. H
e has published research papers in top conferences in cryptography such as
Foundations of Computer Science (FOCS)\, Crypto\, Eurocrypt\, and TCC. Hi
s research has been recognized with the best paper award at Eurocrypt 2018
and invitations to the Journal of Cryptology.\n
URL:https://www.tcs.tifr.res.in/web/events/1047
DTSTART;TZID=Asia/Kolkata:20200127T110000
DTEND;TZID=Asia/Kolkata:20200127T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1048
DTSTAMP:20230914T125948Z
SUMMARY:The Directed Subgraph Homeomorphism Problem
DESCRIPTION:Speaker: Vipin S\n\nAbstract: \nThe Directed Subgraph Homeomor
phism Problem"\, Fortune\, Hopcroft and Wyllie\, Theoretical Computer Scie
nce\, 1980.\n
URL:https://www.tcs.tifr.res.in/web/events/1048
DTSTART;TZID=Asia/Kolkata:20200128T140000
DTEND;TZID=Asia/Kolkata:20200128T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1049
DTSTAMP:20230914T125948Z
SUMMARY:Weight distribution and List-Decoding size of Reed-Muller codes
DESCRIPTION:Speaker: Tulasi mohan Molli\n\nAbstract: \nDetails: We will fin
ish the remainder of previous student talk. We will begin where we ended l
ast time.\nAbstract of the previous talk: The problem of list-decoding an
error-correcting code is the following:\ngiven a received word and a dista
nce parameter find all codewords of the code that are within the given di
stance from the received word. It is a generalization of the more common
notion of unique decoding.\nThe weight distribution of an error-correcting
code counts\, for every given weight parameter\, the number of codewords
whose hamming weight is less than the given weight parameter.\nThe codew
ords of Reed-Muller code can be thought of as truth-tables of low degree
polynomials. Kaufman\, Lovett\, and Porat in their work from 2010 made a
novel connection between computer science techniques used for studying lo
w-degree polynomials and these seemingly related coding theory questions
in the case of Reed-Muller codes.\nIn this talk\, we will see\n1) The abov
e-mentioned result of Kaufman\, Lovett\, and Porat[KLP10] and subsequent
improvement of it due to Abbe\, Sphilka\, and Wigderson[ASW15] which give
upper bounds on the weight distribution of Reed-Muller codes.\n2) A lower
bound on the weight-distribution of Reed-Muller codes by exhibiting a co
llection of low-degree polynomials that satisfy the weight requirement fo
r any given weight parameter. This is joint work with Bhandari\, Harsha\,
and Saptarishi and independently discovered by Sberlo and Shpilka[SS20].
\n
URL:https://www.tcs.tifr.res.in/web/events/1049
DTSTART;TZID=Asia/Kolkata:20200131T140000
DTEND;TZID=Asia/Kolkata:20200131T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1050
DTSTAMP:20230914T125948Z
SUMMARY:Quantum versus Randomized Communication Complexity\, with Efficient
Players
DESCRIPTION:Speaker: Uma Girish (Princeton University\nUnited States)\n\nAb
stract: \nAbstract: We study a new type of separation between quantum a
nd classical communication complexity\, a separation which is obtained usi
ng quantum protocols where all parties are efficient\, in the sense that t
hey can be implemented by small quantum circuits with oracle access to the
ir inputs. More precisely\, we give an explicit partial Boolean function t
hat can be computed in the quantum-simultaneous-with-entanglement model of
communication\, however\, every interactive randomized protocol is of exp
onentially larger cost. Furthermore\, all the parties in the quantum proto
col can be implemented by quantum circuits of small size with blackbox acc
ess to the inputs. Our result qualitatively matches the strongest known se
paration between quantum and classical communication complexity and is obt
ained using a quantum protocol where all parties are efficient. Our proof
technique is new in the context of communication complexity and is based o
n techniques from the recent oracle separation of BQP and PH.\nThis is a j
oint work with Ran Raz and Avishay Tal.\n
URL:https://www.tcs.tifr.res.in/web/events/1050
DTSTART;TZID=Asia/Kolkata:20200207T113000
DTEND;TZID=Asia/Kolkata:20200207T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1051
DTSTAMP:20230914T125948Z
SUMMARY:Artificial Intelligence using Emerging Eevices and Architectures
DESCRIPTION:Speaker: Dr. Debanjan Bhowmik (Assistant Professor\nDepartment
of Electrical Engineering\nIndian Institute of Technology Delhi)\n\nAbstra
ct: \nAbstract: Artificial Intelligence (AI)/ Machine Learning (ML)/ Neur
al Network (NN) algorithms are being widely used currently for various app
lications that include self driving cars\, virtual assistants on smartphon
es and other devices\, etc. However\, the memory-computing separation in e
xisting computer hardware makes implementation of these algorithms on the
hardware inefficient in terms of power and speed. As a result\, new device
s and architectures are being proposed to run these algorithms more effici
ently.\nIn this context\, I will discussed the latest work carried out in
our research group on the implementation of AI/ML/NN algorithms in a cross
bar based in-memory computing architecture\, as well as a quantum architec
ture. For the former\, we have used both spin based devices (spin orbit to
rque driven domain wall devices) [1\,2\,3] and charge based devices (a sin
gle conventional silicon transistor synapse) [4]. We have used both non-sp
iking NN algorithms\, used abundantly in the ML community\, [2\,4] as well
as spiking NN algorithms\, inspired from the working of the brain [3]. Fo
r the latter\, we have proposed a novel quantum algorithm\, implemented it
on the "qiskit" simulation framework and shown very high classification a
ccuracy on different popular ML datasets [5].\nReferences:\n1. Debanjan Bh
owmik et al. "Deterministic domain wall motion orthogonal to current flow
due to spin orbit torque". Scientific Reports \, Vol. 5\, 11823 (2015)\n2.
Debanjan Bhowmik et al. "On-chip learning for domain wall synapse based F
ully Connected Neural Network". Journal of Magnetism and Magnetic Material
s Vol. 489\, 165434 (2019)\n3. Upasana Sahu\, Aadit Pandey\, Kushaagra Goy
al and Debanjan Bhowmik. "Spike time dependent plasticity (STDP) enabled l
earning in spiking neural networks using domain wall based synapses and ne
urons". AIP Advances Vol. 9\, 12 (2019)\n4. Nilabjo Dey\, Janak Sharda\, U
tkarsh Saxena\, Divya Kaushik\, Utkarsh Singh\, Debanjan Bhowmik. "On-Chip
Learning in a Conventional Silicon MOSFET Based Analog Hardware Neural Ne
twork". IEEE Biomedical Circuits and Systems Conference (BioCAS)\, Nara\,
Japan (2019)\n5. S. Adhikary\, S. Dangwal and D. Bhowmik\, Supervised lear
ning with a quantum classifier using multi-level systems\, Quantum Informa
tion Processing 19\, 89 (2020).\nBio: Dr. Debanjan Bhowmik is currently a
n Assistant Professor in the Department of Electrical Engineering\, Indian
Institute of Technology Delhi. He obtained his BTech degree in Electrical
Engineering from Indian Institute of Technology Kharagpur in 2010. He obt
ained his PhD degree from University of California Berkeley in 2015\, work
ing in the field of nano magnetism and spintronics. Currently at IIT Delhi
he works on Artificial Intelligence using emerging devices like spintroni
c devices and emerging architectures like in-memory computing architecture
and quantum architecture.\n
URL:https://www.tcs.tifr.res.in/web/events/1051
DTSTART;TZID=Asia/Kolkata:20200211T140000
DTEND;TZID=Asia/Kolkata:20200211T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1052
DTSTAMP:20230914T125948Z
SUMMARY:Characterization of Reed Muller Codes
DESCRIPTION:Speaker: Arghya Chakraborty\n\nAbstract: \nAbstract : In this t
alk\, I shall discuss about linear codes-specifically Reed Solomon\, Reed
Muller codes and their duals. Reed Muller codes are a generalization of Re
ed Solomon codes but here we will also see that Reed Muller is a subcode o
f Reed Solomon codes.\n
URL:https://www.tcs.tifr.res.in/web/events/1052
DTSTART;TZID=Asia/Kolkata:20200212T160000
DTEND;TZID=Asia/Kolkata:20200212T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1053
DTSTAMP:20230914T125948Z
SUMMARY:The Propp-Wilson Perfect Sampling Algorithm
DESCRIPTION:Speaker: Sayantan Chakraborty (TIFR\, Mumbai)\n\nAbstract: \nAb
stract:*The problem of approximate sampling using Markov Chain Monte Carlo
has received considerable attention in the Theoretical Computer Science a
nd Physics communities. In this approach one typically takes a Markov Chai
n $M$ with a stationary distribution $\\Pi$ over a set $\\mathcal{S }$ and
runs it for a pre-specified amount of time $T$ from a fixed starting stat
e $s_0$. Using various techniques such as Coupling\, Spectral Gap\, etc.\,
the distribution of the output of the chain after $T$ steps\, say\, $s_T$
\, is shown to be close to the stationary distribution $\\Pi$ in total v
ariation distance . $T$ typically has a dependence on $\\log(1/\\eps)$ whe
re $\\eps$ is the desired upper bound on the TV distance between the distr
ibution of $s_T$ and $\\Pi$.* *However\, in this talk we will be intereste
d in producing a sample from $\\mathcal{S}$ that is distributed exactly ac
cording to $\\Pi$\, i.e\, we require $\\eps=0$. Notice\, that the above ap
proximate sampling approach cannot be used directly as we would require $T
$ to be $\\infty$ as it depends on $\\log(1/\\eps)$. To overcome the above
issues Propp and Wilson [1996] came up with a beautiful algorithm known a
s Coupling From The Past (CFTP). In this talk we will arrive at the algori
thm by taking various erroneous approaches and correcting them.*\n
URL:https://www.tcs.tifr.res.in/web/events/1053
DTSTART;TZID=Asia/Kolkata:20200214T140000
DTEND;TZID=Asia/Kolkata:20200214T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1054
DTSTAMP:20230914T125948Z
SUMMARY:Unique Games with Value Half and its Applications.
DESCRIPTION:Speaker: Amey Bhangale (University of California\, Riverside\nC
alifornia\, U.S.)\n\nAbstract: \nAbstract: PCP Theorem characterizes the c
lass NP and gives hardness of approximation for many optimization problems
. Despite this\, several tight hardness of approximation results remain el
usive via the PCP theorem. In 2002\, Khot proposed the Unique Games Conjec
ture. Since the formulation of the conjecture\, it has found interesting c
onnections to the tight hardness of approximation results for many optimiz
ation problems. \n\nIn this talk\, I will discuss recent developments abou
t the Unique Games Conjecture. The 2-to-2 Games Theorem implies that it is
NP-hard to distinguish between Unique Games instances with assignment sat
isfying at least 1/2 fraction of the constraints vs. no assignment satisfy
ing more than \\eps fraction of the constraints\, for every constant \\eps
>0. We show that the reduction can be transformed in a non-trivial way to
give a stronger guarantee on the Unique Games instance and use this guaran
tee to convert the known Unique Games-hardness results to NP-hardness for
several optimization problems. \n\nBased on joint work with SubhashK Khot.
\n
URL:https://www.tcs.tifr.res.in/web/events/1054
DTSTART;TZID=Asia/Kolkata:20200218T141500
DTEND;TZID=Asia/Kolkata:20200218T151500
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1055
DTSTAMP:20230914T125948Z
SUMMARY:The Skolem - Mahler - Lech Theorem
DESCRIPTION:Speaker: Prabhat Kumar Jha\n\nAbstract: \nAbstract: The Skolem
- Mahler - Lech Theorem states that given any linear recurrence sequence o
ver any field of characteristic 0\, the set of positions where 0 occurs is
union of a finite set and finitely many arithmetic progressions. We will
discuss a proof of this theorem.\n
URL:https://www.tcs.tifr.res.in/web/events/1055
DTSTART;TZID=Asia/Kolkata:20200221T140000
DTEND;TZID=Asia/Kolkata:20200221T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1056
DTSTAMP:20230914T125948Z
SUMMARY:Optimal Resource Allocation and Scheduling in Networks and Systems
DESCRIPTION:Speaker: Rahul Singh (The Ohio State University\nColumbus\, Ohi
o\, United States.)\n\nAbstract: \nAbstract: I will provide an account of
my research in various areas. As an exemplar\, I will discuss the first tw
o topics in greater detail.\n1) Decentralized Control of Stochastic Dynami
cal Systems: We begin by developing new methods to design decentralized c
ontrol laws for stochastic dynamical systems which perform as well as an o
ptimal centralized policy. We illustrate these methods on real-time multi
-hop communication networks. The problem is challenging because it involv
es inducing coordination amongst the controllers without knowing all the s
tates of individual agents.\n\n2) Reinforcement Learning: We consider the
problem of designing learning rules for Markov decision processes under c
onstraints on the cost expenditures by the controller/agent.\n\n3) Asympto
tic Smoothness of a Service Discipline: We introduce a new performance met
ric that is useful in order to characterize schedulers for networks servi
ng real-time traffic. We also show that the popular MaxWeight scheduler p
erforms well with respect to it.\n\n4) Networked Control Systems: We addre
ss the problem of how to optimally schedule data packets over an unreliabl
e channel in order to minimize the estimation error of a remote linear es
timator that tracks the state of a Gauss Markov process. We show that a s
imple index rule that calculates the value of information (VoI) of each pa
cket\, and then schedules the packet with the largest current value of Vo
I\, is optimal.\n\nI will conclude the talk by discussing my immediate and
long-term research directions.\n\nBio: Rahul Singh is a postdoctoral res
earcher at the Ohio State University. He received the B.Tech. degree in El
ectrical Engineering from Indian Institute of Technology Kanpur in 2009\,
the M.S. degree in Electrical Engineering from the University of Notre D
ame in 2011\, and the Ph.D. degree in Computer Engineering from the Depart
ment of Electrical and Computer Engineering\, Texas A&M University\, Co
llege Station\, in 2015. He was Postdoctoral Researcher at the Laboratory
for Information and Decision Systems (LIDS)\, Massachusetts Institute of
Technology. He also worked as a Data Scientist at Encored Inc.\, and was
part of the Machine Learning Group at Intel. His research interests inclu
de decentralized control of large-scale complex cyber-physical systems\, o
peration of electricity markets with renewable energy\, optimal schedulin
g and control of networks serving real time traffic\, machine learning\,
game theory\, stochastic control\, multi-armed bandits and reinforcement l
earning.\n
URL:https://www.tcs.tifr.res.in/web/events/1056
DTSTART;TZID=Asia/Kolkata:20200224T110000
DTEND;TZID=Asia/Kolkata:20200224T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1057
DTSTAMP:20230914T125948Z
SUMMARY:STCS Annual Symposium
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1057
DTSTART;TZID=Asia/Kolkata:20200228T093000
DTEND;TZID=Asia/Kolkata:20200301T163000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1058
DTSTAMP:20230914T125948Z
SUMMARY:Integer Linear Programs and Local Search for Max-Cut (Author: Svato
pluk Poljak)
DESCRIPTION:Speaker: Vidya Sagar Sharma\n\nAbstract: \nAbstract: Schaffer
and Yannakakis have shown that the max-cut problem with the FLIP neighborh
ood is polynomial-time local search (PLS) complete\, and hence among the m
ost difficult problems in the PLS class. The FLIP neighbourhood of a 2-par
tition is defined by moving a single vertex to the opposite class. In this
paper\, the author has shown that when restricted to the cubic graph\, th
e FLIP local search becomes "easy" and finds a local max-cut in $O(n^2)$ s
teps.\nPaper Link: https://epubs.siam.org/doi/pdf/10.1137/S009753979324535
0\n
URL:https://www.tcs.tifr.res.in/web/events/1058
DTSTART;TZID=Asia/Kolkata:20200306T140000
DTEND;TZID=Asia/Kolkata:20200306T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1059
DTSTAMP:20230921T105045Z
SUMMARY:Distribution oblivious risk-aware bandit algorithms
DESCRIPTION:Speaker: Jayakrishnan Nair (Indian Institute of Technology Bomb
ay)\n\nAbstract: \nClassical multi-armed bandit algorithms are tuned to wo
rk well for a pre-specified class of reward distributions\, defined via th
eir support\, or moment/tail bounds. In this work\, we consider the proble
m of oblivious best arm identification\, i.e.\, where the algorithm has no
prior information about the class of reward distributions. We establish f
undamental limits on the performance of oblivious algorithms\, and further
propose algorithms that asymptotically meet these limits. Additionally\,
we allow for risk-aware arm selection\, where we balance the expected rewa
rd associated with an arm with the corresponding risk.\nThis talk is based
on joint work with Anmol Kagrecha\, Ashutosh Kumar\, and Krishna Jagannat
han.\n
URL:https://www.tcs.tifr.res.in/web/events/1059
DTSTART;TZID=Asia/Kolkata:20200505T160000
DTEND;TZID=Asia/Kolkata:20200505T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1060
DTSTAMP:20230921T105045Z
SUMMARY:Battle of Bandits: Online Learning from Relative Preferences
DESCRIPTION:Speaker: Aditya Gopalan (Indian Institute of Science\, Bangalor
e)\n\nAbstract: \nWe consider the problem of sequentially learning good al
ternatives from among a pool\, but with only relative utility feedback fro
m adaptively chosen subsets. At each round\, the learner chooses a subset
of alternatives and can observe which ones are preferred over the others i
n the subset. This type of feedback is natural in several domains\, especi
ally where human preferences are elicited in a repeated fashion ("Which of
A\, B\, C\, D do you prefer?")\, e.g.\, the design of surveys and expert
reviews\, web search and recommender systems\, and other settings like ran
king in multiplayer games. Tranditional approaches such as the multi-armed
bandit model only absolute utility feedback\, and are thus inadequate to
express relative choices. The dueling bandit (Yue-Joachims'09) is a more r
ecent attempt to model online learning with pairwise preferences\, but the
more general\, realistic\, and combinatorially harder case of working wit
h preferences expressed over subsets has largely been unexplored. We take
a step in this direction and formulate what we call the battling bandit pr
oblem\, where one seeks to learn an optimal item or ranking of n items by
sequentially choosing up to size-k subsets at each round and exploiting re
lative preferences arising from a choice model such as the well-known Plac
kett-Luce probability model. We study variants of learning objectives from
subsetwise feedback: Identifying the best item\, the set of top-k items\,
full ranking etc.\, in both the probably approximately correct (PAC) or r
egret optimization setting\, and design algorithms with optimality propert
ies. This is joint work with Aadirupa Saha (Indian Institute of Science).\
nBio:\nAditya Gopalan is an Assistant Professor and INSPIRE Faculty Fellow
at the Indian Institute of Science\, Dept. of Electrical Communication En
gineering. He received the Ph.D. degree in electrical engineering from The
University of Texas at Austin\, and the B.Tech. and M.Tech. degrees in el
ectrical engineering from the Indian Institute of Technology Madras. He wa
s an Andrew and Erna Viterbi Post-Doctoral Fellow at the Technion-Israel I
nstitute of Technology. His research interests include machine learning an
d statistical inference\, control\, and resource allocation algorithms.\n
URL:https://www.tcs.tifr.res.in/web/events/1060
DTSTART;TZID=Asia/Kolkata:20200512T140000
DTEND;TZID=Asia/Kolkata:20200512T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1061
DTSTAMP:20230914T125949Z
SUMMARY:Tapestry: A Single-Round Smart Pooling Technique for COVID-19 Testi
ng
DESCRIPTION:Speaker: Manoj Gopalkrishnan (Department of Electrical Engineer
ing\nIndian Institute of Technology Bombay\nMumbai)\n\nAbstract: \nAbstrac
t: The COVID-19 pandemic has strained testing capabilities worldwide. Ther
e is an urgent need to find economical and scalable ways to test more peop
le. We present Tapestry\, a novel quantitative nonadaptive pooling scheme
to test many samples using only a few tests. The underlying molecular diag
nostic test is any real-time RT-PCR diagnostic panel approved for the dete
ction of the SARS-CoV-2 virus. In cases where most samples are negative fo
r the virus\, Tapestry accurately identifies the status of each individual
sample with a single round of testing in fewer tests than simple two-roun
d pooling. We also present a companion Android application BYOM Smart Test
ing which guides users through the pipetting steps required to perform the
combinatorial pooling. The results of the pooled tests can be fed into th
e application to recover the status and estimated viral load for each indi
vidual sample.\n
URL:https://www.tcs.tifr.res.in/web/events/1061
DTSTART;TZID=Asia/Kolkata:20200526T140000
DTEND;TZID=Asia/Kolkata:20200526T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1062
DTSTAMP:20230921T105045Z
SUMMARY:Symmetry in Syndrome Decoding of Reed-Muller Codes
DESCRIPTION:Speaker: Andrew Thangaraj (IIT Madras)\n\nAbstract: \nReed--Mul
ler (RM) codes\, a classical family of codes known for their elegant algeb
raic structure\, have recently been shown to achieve capacity under maximu
m-likelihood (ML) decoding on the binary erasure channel and this has reki
ndled interest in their efficient decoding. We consider the code family RM
(m-3\,m) and develop a new ML decoder\, for transmission over the binary s
ymmetric channel\, that exploits their large symmetry group. The new decod
er has complexity O(m^3)\, which is lower than an earlier method introduce
d by Seroussi and Lempel in 1983.\nBio: Andrew Thangaraj received his B.Te
ch in Electrical Engineering from the Indian Institute of Technology (IIT)
\, Madras\, India in 1998 and a PhD in Electrical Engineering from the Geo
rgia Institute of Technology\, Atlanta\, USA in 2003. He was a post-doctor
al researcher at the GTL-CNRS Telecom lab at Georgia Tech Lorraine\, Metz\
, France from August 2003 to May 2004. From June 2004\, he has been with t
he Department of Electrical Engineering\, IIT Madras\, where he is current
ly a professor. From Jan 2012 to Jan 2018\, he served as Editor for the IE
EE Transactions on Communications.\nSince July 2018\, he has been serving
as an Associate Editor for Coding Techniques for the IEEE Transactions on
Information Theory. Since Oct 2011\, he has been serving as NPTEL coordina
tor at IIT Madras. He has played a key role in initiating and running NPTE
L online courses and certification. He is currently a National MOOCs Coord
inator for NPTEL in the SWAYAM project of the MHRD.\n
URL:https://www.tcs.tifr.res.in/web/events/1062
DTSTART;TZID=Asia/Kolkata:20200602T140000
DTEND;TZID=Asia/Kolkata:20200602T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1063
DTSTAMP:20230914T125949Z
SUMMARY:Weighted min-cut: Sequential\, Cut-query and Streaming algorithms
DESCRIPTION:Speaker: Dr. Sagnik Mukhopadhyay (KTH\, Stockholm)\n\nAbstract:
\nAbstract: Consider the following 2-respecting min-cut problem: Given a
weighted graph G and its spanning tree T\, find the minimum cut among the
cuts that contain at most two edges in T. This problem is an important sub
routine in Karger's celebrated randomized near-linear-time min-cut algorit
hm [STOC'96]. I will present a new approach to this problem which can be e
asily implemented in many settings\, leading to the following randomized m
in-cut algorithms for weighted graphs.\n- An O(m log^2 n)-time sequential
algorithm: This improves Karger's long-standing O(m log^3 n) bound. Improv
ements over Karger's bounds were previously known only under a rather stro
ng assumption that the input graph is simple (unweighted without parallel
edges) [Henzinger\, Rao\, Wang\, SODA'17\; Ghaffari\, Nowicki\, Thorup\, S
ODA'20]. For unweighted graphs (possibly with parallel edges) and using bi
t operations\, our bound can be further improved.\n- An algorithm that req
uires \\tilde O(n) cut queries to compute the min-cut of a weighted graph:
This answers an open problem by Rubinstein\, Schramm\, and Weinberg [ITCS
'18]\, who obtained a similar bound for simple graphs. Our bound is tight
up to polylogarithmic factors.\n- A dynamic streaming algorithm that requi
res \\tilde O(n) space and O(log n) passes to compute the min-cut: The onl
y previous non-trivial exact min-cut algorithm in this setting is the 2-pa
ss \\tilde O(n)-space algorithm on simple graphs [Rubinstein et al.\, ITCS
'18] (observed by Assadi\, Chen\, and Khanna [STOC'19]).\nIn contrast to K
arger's 2-respecting min-cut algorithm which deploys sophisticated dynamic
programming techniques\, our approach exploits some cute structural prope
rties so that it only needs to compute the values of \\tilde O(n) cuts cor
responding to removing \\tilde O(n) pairs of tree edges\, an operation tha
t can be done quickly in many settings.\nThis is joint work with Danupon N
anongkai.\n
URL:https://www.tcs.tifr.res.in/web/events/1063
DTSTART;TZID=Asia/Kolkata:20200609T140000
DTEND;TZID=Asia/Kolkata:20200609T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1064
DTSTAMP:20230914T125949Z
SUMMARY:Optimal Design of Queuing Systems via Compositional Stochastic Prog
ramming
DESCRIPTION:Speaker: Ketan Rajawat (Department of Electrical Engineering\,\
nIndian Institute of Technology\, Kanpur)\n\nAbstract: \nAbstract: Well-de
signed queuing systems form the backbone of modern communications\, distri
buted computing\, and content delivery architectures. Designs balancing in
frastructure costs and user experience indices require tools from tele-tra
ffic theory and operations research. A standard approach to designing such
systems involves formulating optimization problems that strive to maximiz
e the pertinent utility functions while adhering to quality-of-service and
other physical constraints. In many cases\, formulating such problems nec
essitates making simplistic assumptions on arrival and departure processes
to keep the problem tractable.\n\nThis talks will introduce a stochastic
optimization framework for designing queuing systems where the exogenous p
rocesses may have arbitrary and unknown distributions. We show that many s
uch queuing design problems can generally be formulated as stochastic opti
mization problems where the objective and constraints are non-linear funct
ions of expectations. The compositional structure obviates the use of clas
sical stochastic approximation approaches where the stochastic gradients a
re often required to be unbiased. To this end\, a constrained stochastic c
ompositional gradient descent algorithm is proposed that utilizes a tracki
ng step for the expected value functions. The non-asymptotic performance o
f the proposed algorithm is characterized by its iteration complexity. Fur
ther improvements are proposed that build upon the primal-dual saddle poin
t algorithm to result in zero constraint violation and O(T-0.25) optimalit
y gap. Numerical tests allow us to validate the theoretical results and de
monstrate the efficacy of the proposed algorithm.\n\nBio: Ketan Rajawat (S
'06–M'12) received his B.Tech and M.Tech degrees in Electrical Engineeri
ng from the Indian Institute of Technology (IIT) Kanpur\, India\, in 2007\
, and his Ph.D. degree in Electrical and Computer Engineering from the Uni
versity of Minnesota\, Minneapolis\, MN\, USA\, in 2012. He is currently a
n Associate Professor in the Department of Electrical Engineering\, IIT Ka
npur. His research interests are in the broad areas of signal processing\,
robotics\, and communications networks\, with particular emphasis on dist
ributed optimization and online learning. His current research focuses on
the development and analysis of distributed and asynchronous optimization
algorithms\, online convex optimization algorithms\, stochastic optimizati
on algorithms\, and the application of these algorithms to problems in mac
hine learning\, communications\, and smart grid systems. He is currently s
erving as an Associate Editor with the IEEE Communications Letters and IEE
E Transactions on Signal Processing. He is also the recipient of the 2018
INSA Medal for Young Scientists and the 2019 INAE Young Engineer Award.\n
URL:https://www.tcs.tifr.res.in/web/events/1064
DTSTART;TZID=Asia/Kolkata:20200616T140000
DTEND;TZID=Asia/Kolkata:20200616T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1065
DTSTAMP:20230914T125949Z
SUMMARY:Half-integral Duals\, Connectivity Augmentation and Multiflows in P
lanar Graphs
DESCRIPTION:Speaker: Naveen Garg (Computer Science and Engineering\nIndian
Institute of Technology Delhi\nNew Delhi)\n\nAbstract: \nAbstract: Given a
n edge weighted graph and a spanning tree\, $T$\, the {\\em tree augmentat
ion problem} is to pick a minimum weight set of edges\, $F$\, such that $T
\\cup F$ is 2-edge connected. Williamson et.al. gave a 2-approximation alg
orithm for this problem using the primal-dual schema. We show that when ed
ge weights are integral\, the WGMV procedure can be modified to obtain a h
alf-integral dual.\n\nThe tree augmentation problem has an interesting con
nection to routing flow in graphs where the union of supply and demand is
planar. The half-integrality of the dual leads to a tight 2-approximate ma
x-half-integral-flow min-multicut theorem and a 4-approximate max-integral
flow min-multicut theorem.\n(joint work with Nikhil Kumar and Andras Sebo
)\n
URL:https://www.tcs.tifr.res.in/web/events/1065
DTSTART;TZID=Asia/Kolkata:20200623T140000
DTEND;TZID=Asia/Kolkata:20200623T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1066
DTSTAMP:20230914T125949Z
SUMMARY:Proof Complexity of MaxSAT Resolution
DESCRIPTION:Speaker: Gaurav Sood (Institute of Mathematical Sciences\nChenn
ai\, Tamil Nadu)\n\nAbstract: \nZoom details:\nMeeting ID: 613 566 8340\nP
assword: 183102\nAbstract: Boolean satisfiability (SAT) is the quintessent
ial NP-complete problem. It has given rise to many areas of research wit
hin theoretical computer science\, one of them being proof complexity. Gi
ven a proof system for SAT\, we ask questions of the following form: is th
ere a short proof that an unsatisfiable formula is unsatisfiable?\n\nMaxSA
T is the related problem of determining the maximum number of satisfiable
clauses\, and like SAT\, it has its own proof systems. In this talk\, we
will study a proof system for MaxSAT proposed by Bonet et al. in 2007\, an
d compare it with proof systems for SAT. This is joint work with Yuval Fil
mus\, Meena Mahajan and Marc Vinyals.\n
URL:https://www.tcs.tifr.res.in/web/events/1066
DTSTART;TZID=Asia/Kolkata:20200625T143000
DTEND;TZID=Asia/Kolkata:20200625T153000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1067
DTSTAMP:20230914T125949Z
SUMMARY:Theory of Field $\\mathbb{C} $ is Decidable
DESCRIPTION:Speaker: Prabhat Kumar Jha\n\nAbstract: \nZoom Details:\nLink:
https://zoom.us/j/98172222418?pwd=b0htSDQ5NDRKQ050K0d6cHJ3YnZXQT09\nMeetin
g ID: 981 7222 2418\nPassword: studsem\nAbstract: Many optimization pro
blems reduce to finding truth value of quantified formulas over real or/an
d complex numbers. In this talk we will cover a proof of decidability of t
heories of real numbers and complex numbers due to Tarski. We will begin b
y providing an algorithm to locate roots of multi-variable polynomials ove
r these fields and then use quantifier elimination techniques to prove dec
idability. If time permits\, we will discuss its applications to optimizat
ion and verification problems.\nNote: There is no prerequisite from Mathem
atical Logic or Computability theory.\n
URL:https://www.tcs.tifr.res.in/web/events/1067
DTSTART;TZID=Asia/Kolkata:20200704T160000
DTEND;TZID=Asia/Kolkata:20200704T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1068
DTSTAMP:20230914T125949Z
SUMMARY:On Decision Tree Complexity of Boolean Functions
DESCRIPTION:Speaker: Swagato Sanyal (Indian Institute of Technology Kharagp
ur\nKharagpur\, West Bengal)\n\nAbstract: \nTalk will be held on Zoom.\nAb
stract: The decision tree model\, also known as the query model\, is a sim
ple model of computation. Besides having an intuitive appeal\, and offerin
g a natural paradigm for algorithm design\, it is often possible to theore
tically nail down the exact complexity of many interesting algorithmic tas
ks using currently available methods\, completely and unconditionally\; fo
r more complex models of computation such an aspiration is currently a dis
tant dream. In this talk we will first introduce and motivate query model\
, and then present a recent work of ours on the query complexity of an imp
ortant subclass of problems\, called composed functions. The latter part i
s based on joint work with Dmitry Gavinsky\, Troy Lee and Miklos Santha wh
ich has appeared in ICALP 2019.\n
URL:https://www.tcs.tifr.res.in/web/events/1068
DTSTART;TZID=Asia/Kolkata:20200707T140000
DTEND;TZID=Asia/Kolkata:20200707T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1069
DTSTAMP:20230914T125949Z
SUMMARY:On Realizable Distributions
DESCRIPTION:Speaker: Siddharth Bhandari\n\nAbstract: \nJoin Zoom Meeting\nh
ttps://zoom.us/j/98172222418?pwd=b0htSDQ5NDRKQ050K0d6cHJ3YnZXQT09\nMeeting
ID: 981 7222 2418\nPassword: studsem\nAbstract: Coupling arguments are co
mmonplace in approximate and perfect sampling. A coupling is a joint distr
ibution between RVs: it helps us relate and compare distributions of the R
Vs.\n\nWe will study the following:\nGiven RVs X_1\, X_2\, X_3\, X_4\,....
\, X_n each taking a value in $\\Omega$ and marginally distributed accordi
ng to $\\mu_1\, \\mu_2\, ...\, \\mu_n$ on $\\Omega^n$\, we say a RV $Y$ di
stributed according to $\\tau$ is realizable wrt to $X_1\,...\, X_n$ iff f
or all joint distributions $\\mu$ of $X_1\,..\, X_n$ such that each $X_i~\
\mu_i$ there exists a joint distribution $\\mu'$ of $X_1\,...\, X_n\, Y$
such that 1> $(X_1\, X_2\, ...\, X_n)~\\mu$ and $Y~\\tau$ 2> Pr_{\\mu
'}[Y\\in {X_1\,...\, X_n}]=1.\nWe will show that a distribution $\\tau$
is realizable wrt to $\\mu_1\,..\, \\mu_n$ iff for all $S\\subseteq \\Omeg
a$ we have: $\\mu_i(S)\\leq \\tau(S)\\leq \\mu_j (S)$ for some $i\,j\\in [
n]$.\n
URL:https://www.tcs.tifr.res.in/web/events/1069
DTSTART;TZID=Asia/Kolkata:20200711T160000
DTEND;TZID=Asia/Kolkata:20200711T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1070
DTSTAMP:20230914T125949Z
SUMMARY:Approximating the Nash Social Welfare with Subadditive Valuations
DESCRIPTION:Speaker: Umang Bhaskar\n\nAbstract: \nAbstract: How should o
ne divide $m$ goods between $n$ agents\, given the utility each agent has
when allocated a subset of goods? Allocations which maximize the Nash Soci
al Welfare --- the product of agent utilities for their allocation --- are
known to possess a number of natural and agreeable properties. Unfortunat
ely\, maximizing the Nash Social Welfare is known to be APX-hard\, even wh
en agent utility functions are additive over the set of goods.\nWe present
an algorithm that for the general case of subadditive utility functions f
or the agents\, obtains an $O(n)$-approximation to the optimal Nash Social
Welfare\, given value oracle access to the utility functions. This improv
es upon a previous $O(n log n)$ algorithm for submodular utilities (SODA '
20). Our approximation ratio is tight for subadditive utilities and for va
lue oracle access.\nThis is joint work with Siddharth Barman\, Anand Krish
na\, and Ranjani Sundaram.\n
URL:https://www.tcs.tifr.res.in/web/events/1070
DTSTART;TZID=Asia/Kolkata:20200717T140000
DTEND;TZID=Asia/Kolkata:20200717T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1071
DTSTAMP:20230914T125949Z
SUMMARY:The class of poly-sized Algebraic Branching Programs is closed unde
r Factoring
DESCRIPTION:Speaker: Prerona Chatterjee\n\nAbstract: \nTalk will be held i
n Zoom.\nAbstract: In the 1980s\, Kaltofen proved one of the most remar
kable results in algebraic complexity theory. He showed that if a polynomi
al can be computed by a "small" circuit\, then each of its factors can als
o be computed by "small" circuits. In fact\, given a circuit for the origi
nal polynomial\, he also gave an efficient algorithm for computing circuit
s for the factors. This result has many applications\, one of which is the
algebraic analogue of the "hardness vs randomness" question.\n\nHowever\,
his proof did not extend to more restricted models of computation. Recent
ly Amit Sinhababu and Thomas Thierauf showed a similar statement for the c
lass of Algebraic Branching Programs (or ABPs). Formally\, they proved the
following statement. If an n variate degree d polynomial f can be compute
d by an ABP of size s\, then each of its factors can be computed by an ABP
of size at most poly(s\, n\, d).\nIn this talk\, we will go over the proo
f techniques used in the previous results of this kind\, and see an overvi
ew of the modifications Sinhababu and Theirauf made to these techniques to
get their result.\n\nNo background of the field will be assumed.\n
URL:https://www.tcs.tifr.res.in/web/events/1071
DTSTART;TZID=Asia/Kolkata:20200718T160000
DTEND;TZID=Asia/Kolkata:20200718T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1072
DTSTAMP:20230914T125949Z
SUMMARY:A Largish Sum-of-squares Implies Circuit Hardness (and Derandomizat
ion)
DESCRIPTION:Speaker: Prof. Nitin Saxena (Indian Institute of Technology Kan
pur\nKanpur\, Uttar Pradesh)\n\nAbstract: \nAbstract: We study the sum of
squares (*SOS*) representation of polynomials\, i.e. f = \\sum_{i\\in s}
c_i f_i^2 \, where c_i are field elements and f_i(x_1\,\\ldots\,x_n) are p
olynomials. We are interested in `measuring' the number of monomials that
appear across f_i's\; call this *support-sum* S(f). Let the degree of f be
d and\, for simplicity of exposition\, fix n=1. We conjecture: For some *
explicit* f and some constant \\epsilon > 1/2\, S(f) \\ge d^\\epsilon. We
prove that the conjecture implies VP\\ne VNP (& blackbox-PIT in QuasiP). A
more sophisticated version of the conjecture\, for sum-of-cubes represent
ation\, implies blackbox-PIT is in P.\n\nNote that S(f) \\ge d^{1/2} is tr
ivially achieved. So\, compared to this\, a marginally better lower bound
in the SOS suffices to solve the major questions of algebraic complexity.
The *SOS-conjecture* is largely field independent\, and is itself supporte
d by a dimension-based fact: It holds for a `randomly chosen' degree-d pol
ynomial f(x).\nThis is ongoing work with Pranjal Dutta (CMI/IITK) and Thom
as Thierauf (Ulm/Aalen).\n
URL:https://www.tcs.tifr.res.in/web/events/1072
DTSTART;TZID=Asia/Kolkata:20200723T140000
DTEND;TZID=Asia/Kolkata:20200723T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1073
DTSTAMP:20230914T125949Z
SUMMARY:Simple\, Credible\, and Approximately-Optimal Auctions
DESCRIPTION:Speaker: Santhoshini Velusamy (Harvard University\nCambridge\,
Massachusetts.)\n\nAbstract: \nAbstract: We present a general framework
\, applicable to both truthful and non-truthful auctions\, for designing a
pproximately revenue-optimal mechanisms for multi-item additive auctions.
Given a (not necessarily truthful) single-item auction format satisfying c
ertain technical conditions\, we run simultaneous item auctions augmented
with a personalized entry fee that each bidder must pay before accessing t
he auction. The entry fee depends only on the prior distribution of bidder
types\, and in particular\, is independent of the realized bids.\nWe boun
d the revenue of the resulting two-part tariff mechanism using a novel geo
metric lemma that enables us to provide revenue guarantees for many common
non-truthful auctions that previously had none. Our framework can be used
with many common auction formats\, such as simultaneous first-price\, sim
ultaneous second-price\, and simultaneous all-pay auctions. For all-pay au
ctions\, we prove that the resulting mechanism is also credible in the sen
se that the auctioneer cannot benefit by deviating from the stated mechani
sm after observing agent bids. This is the first static credible mechanism
for multi-item additive auctions that guarantees a constant factor of the
optimal revenue.\nA paper based on this joint work with Costis Daskalakis
\, Maxwell Fishelson\, Brendan Lucier\, and Vasilis Syrgkanis was presente
d at the Twenty-First ACM Conference on Economics and Computation (EC 2020
). The arXiv version of the paper is available at https://arxiv.org/abs/20
02.06702\n
URL:https://www.tcs.tifr.res.in/web/events/1073
DTSTART;TZID=Asia/Kolkata:20200728T140000
DTEND;TZID=Asia/Kolkata:20200728T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1074
DTSTAMP:20230914T125949Z
SUMMARY:Average Sensitivity of Graph Algorithms
DESCRIPTION:Speaker: Nithin Varma (University of Haifa\, Israel)\n\nAbstrac
t: \nAbstract: In modern applications of graph algorithms\, where the gr
aphs of interest are large and dynamic\, it is unrealistic to assume that
an input representation contains the full information of a graph being stu
died. For example\, consider a social network\, where a vertex corresponds
to a user and an edge corresponds to a friendship relation.\n\nIt is reas
onable to assume that users do not always update new friendships on the so
cial network\, and that sometimes they do not fully disclose their friends
hip relations for security or privacy reasons.\n\nThis motivates the desig
n of graph algorithms that\, in spite of being given only a (large) subgra
ph as input\, output solutions that are close to the solutions output when
the whole graph is available. In this talk\, I will introduce a notion of
sensitivity of graph algorithms that formalizes this desirable feature. A
fter discussing the basic properties of our sensitivity definition\, I wil
l give an overview of our main results\, and present the key ideas used in
the design of our algorithm with low sensitivity for the global minimum c
ut problem.\n\nThis is a joint work with Yuichi Yoshida.\n
URL:https://www.tcs.tifr.res.in/web/events/1074
DTSTART;TZID=Asia/Kolkata:20200808T160000
DTEND;TZID=Asia/Kolkata:20200808T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1075
DTSTAMP:20230914T125949Z
SUMMARY:Coordination Over Networks: Shared Randomness\, Security\, and Inte
raction
DESCRIPTION:Speaker: Gowtham Kuri (TIFR)\n\nAbstract: \nAbstract-In a coord
ination problem\, users in a network observing correlated inputs collabora
te to evaluate possibly randomized functions of the inputs. Problems of th
is kind have been widely studied in the information theory literature. It
finds applications in several diverse areas such as in parallel processing
\, cooperative game theory\, distributed control\, function computation in
networks. The focus of many of the works in the literature has been on th
e amount of communication needed to achieve coordination. In practice\, se
veral other aspects are also of interest such as the amount and form of sh
ared/correlated randomness available\, topology of the network used\, and
security. \nIn this talk\, we present a systematic study of various suc
h aspects\, namely\, shared randomness\, security\, and interaction in add
ition to the amount of communication needed. To this end\, the first probl
em we study is a distributed sampling problem where a set of processors wa
nt to output correlated sequences of random variables with the help of a c
oordinator which has access to several independent sources of randomness a
nd each processor has access to a subset of these sources. We characterize
optimal communication and/or shared randomness rates in various cases of
this setting.\nIn the second problem\, we study interactive secure functio
n computation. The privacy requirement is that the communication should no
t reveal to either user any extra information about the other user’s inp
ut and output other than what can be inferred from the user’s own input
and output. We give single-letter expressions for the asymptotic rate regi
ons. Further\, we analyse the role of common randomness and interaction. I
n secure function computation\, in some settings\, it might be reasonable
to give away a certain amount of information about the inputs but not abou
t some specific functions of the inputs which the users want to keep priva
te. We study such settings also.\n
URL:https://www.tcs.tifr.res.in/web/events/1075
DTSTART;TZID=Asia/Kolkata:20200810T103000
DTEND;TZID=Asia/Kolkata:20200810T113000
LOCATION:Join Zoom Meeting https://zoom.us/j/98083276706?pwd=cm85dHNQU0tSKz
c1SndzYlZKL3ZxZz09 Meeting ID: 980 8327 6706 Passcode: 468237
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1076
DTSTAMP:20230914T125949Z
SUMMARY:Identifying some necessary conditions for a function to be Boolean
DESCRIPTION:Speaker: Nikhil Mande (Georgetown University\, Washington\, D.C
.)\n\nAbstract: \nThe seminar will happen via Google meet.\n\nAbstract: E
very Boolean function $f : \\mathbb{F}_2^n \\to \\{-1\, 1\\}$ has a unique
Fourier representation\, that is\, $f = \\sum_{\\alpha \\in \\mathbb{F}_2
^n}\\hat{f}(\\alpha)\\chi_{\\alpha}$. One can obtain the following identit
ies using elementary techniques.\n\n1) Parseval's identity: $\\sum_{\\alph
a \\in \\mathbb{F}_2^n}\\hat{f}(\\alpha)^2 = 1$.\n2) For all $\\gamma \\ne
q 0^n$\, we have $\\sum_{(\\alpha_1\, \\alpha_2) \\in \\mathbb{F}_2^n \\ti
mes \\mathbb{F}_2^n : \\alpha_1+\\alpha_2=\\gamma}\\hat{f}(\\alpha_1)\\hat
{f}(\\alpha_2)=0$. Henceforth\, we refer to these conditions as Titsworth'
s conditions.\n\nLet $\\mathcal{S}$ denote the Fourier support of $f$\, th
at is\, the set of elements of $\\mathbb{F}_2^n$ that have non-zero Fourie
r coefficients. Titsworth's conditions can be seen to imply that for every
pair $(\\alpha_1\, \\alpha_2)$ of elements in $\\mathcal{S}$\, there must
exist at least one other pair $(\\beta_1\, \\beta_2)$ of elements in $\\m
athcal{S}$ such that $\\alpha_1 \\oplus \\alpha_2 = \\beta_1 \\oplus \\bet
a_2$. We investigate whether this condition in itself is sufficient for $\
\mathcal{S}$ to be the Fourier support of a Boolean function.\n\nOnly basi
c mathematical familiarity will be assumed. I will cover the required prer
equisites from Fourier analysis of Boolean functions in the beginning of t
he talk.\n\nBased on joint work with Swagato Sanyal (IIT Kharagpur). Link
to full paper: https://arxiv.org/abs/2008.00266\n
URL:https://www.tcs.tifr.res.in/web/events/1076
DTSTART;TZID=Asia/Kolkata:20200815T110000
DTEND;TZID=Asia/Kolkata:20200815T120000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1077
DTSTAMP:20230914T125949Z
SUMMARY:Guarding Polygons via CSPs
DESCRIPTION:Speaker: Akanksha Agrawal (Ben-Gurion University\, Israel)\n\nA
bstract: \nAbstract: Introduced by Klee in 1973\, the Art Gallery problem
asks the following question. Given the top view of an art gallery in the f
orm of a polygon\, how many guards does one need to place inside the art g
allery so that each wall is visible to at least one of them? It is easy to
see that for convex polygons\, one guard is always sufficient. The proble
m becomes nontrivial when the polygon has reflex vertices (vertices whose
interior angle is more than 180 degrees).\n\nIn this talk\, we study some
variants of the Art Gallery problem. Using structural properties of almost
convex polygons\, we reduce an Art Gallery variant to a CSP (Constraint S
atisfaction Problem) whose constraints have arity two and involve monotone
functions. We then obtain a polynomial-time algorithm for the CSP\, and c
onsequently for the Art Gallery variant when parameterized by the number o
f reflex vertices. This is joint work with Kristine Knudsen\, Daniel Loksh
tanov\, Saket Saurabh\, and Meirav Zehavi.\n\n(A condensed version of this
talk was presented at the TCS Women Spotlight Workshop at STOC 2020.)\n
URL:https://www.tcs.tifr.res.in/web/events/1077
DTSTART;TZID=Asia/Kolkata:20200821T171500
DTEND;TZID=Asia/Kolkata:20200821T181500
LOCATION:Zoom link: https://zoom.us/j/92834710011?pwd=WURSa3JsRVpJYWhONEExY
TdDcENDQT09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1078
DTSTAMP:20230914T125949Z
SUMMARY:Communication Complexity and Quantum Machine Learning via Query Com
plexity
DESCRIPTION:Speaker: Suhail Sherif\n\nAbstract: \nQuery algorithms are ever
ywhere. Gradient Descent is a well-known algorithm that queries the gradie
nt of a hidden function and moves towards the minimum of the function. Sho
r's factoring algorithm is based on a quantum query algorithm for period f
inding. Many interesting computational tasks are indeed query tasks. Howev
er\, the study of query complexity of total functions also leads us to int
eresting mathematical insights\, such as the degree of a function being po
lynomially related to its query complexity.\n\nIn this thesis we show a fe
w related results.\n\n1. In the context of randomized communication comple
xity and randomized parity decision tree (RPDT) complexity\, we prove that
the analogues of the mathematical insight stated above do not hold. This
disproves multiple long-standing conjectures in communication complexity.\
n\n2. Our results above originated in the RPDT world and were then transla
ted to the communication world. However\, we also have stronger results in
the RPDT world that we could not translate yet. We pose a fundamental con
jecture that would imply that the stronger results do hold in the communic
ation world as well.\n\n3. We analyze the query complexity of minimizing a
convex function in a bounded region when given access to function value a
nd gradient oracles. Projected Gradient Descent is known to be optimal for
deterministic and randomized query algorithms (although the randomized lo
wer bound is for a limited range of parameters). We provide a randomized l
ower bound that works for the optimal range of parameters. We also prove t
hat projected gradient descent is optimal among quantum query algorithms.\
n
URL:https://www.tcs.tifr.res.in/web/events/1078
DTSTART;TZID=Asia/Kolkata:20200824T103000
DTEND;TZID=Asia/Kolkata:20200824T113000
LOCATION:Zoom link for the meeting. https://zoom.us/j/96554737206?pwd=R
zBsVE56TnV2dTZ0dlNmaFp5MFRzZz09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1079
DTSTAMP:20230914T125949Z
SUMMARY:An Asymptotic Study of Financial Systems - Algorithms and Analysis
DESCRIPTION:Speaker: Anand Deo\n\nAbstract: \nOver the past few decades\, a
symptotic study of financial systems has become an integral part of practi
cal decision making and analytics. Realistic financial systems are complex
\, and undesirable events in them are often rare. In order to gain more in
sights into their behaviour\, it is important to develop structural simpli
fications and efficient computational algorithms for rare event analysis.
In this talk\, we undertake a detailed study of these aspects\, and develo
p structural insights on a number of financial systems of practical intere
st.\n\nIn the first part\, we consider two problems related to rare event
analysis:\n\n1) The estimation of default probabilities of financial firms
from data is an important problem which has received significant attentio
n over the past two decades. We discuss the development of a closed form\,
interpretable parameter estimation technique for predicting defaults of f
inancial firms. Typically\, one uses Maximum Likelihood Estimation (MLE) f
or predicting the firm default probabilities. We prove that our estimator
is almost as accurate as the MLE for a realistic sample of financial data.
Further since our estimator is closed form\, it is significantly faster t
han MLE. Finally\, we demonstrate that unlike the MLE our estimator also g
ives interesting structural insights - specifically\, we show that from th
e standpoint of default prediction\, collecting covariate data just before
occurrence of default is sufficient to estimate probabilities.\n\n2) Buil
ding upon the well established notions of multivariate regular variation a
nd large deviations theory\, we derive a unified framework\, in which tail
analysis of a large class of stochastic loss functions can be performed.
Within this framework\, assuming the underlying stochasticity is heavy tai
led\, we develop a data-driven estimator for tail exceedences of a large c
lass of financial losses\, and applying it to a tail risk-constrained port
folio optimisation problem\, showcase superior performance over the state
of the art. Additionally\, assuming oracle access to the densities of loss
causing covariates\, we develop a self-replicating\, provably accurate im
portance sampling algorithm to estimate rare event probabilities over a va
riety of covariate/loss structures.\n\nIn the latter part of the talk\, we
develop a limiting representation for an interconnected banking network i
n presence of partial information. Practical banking networks are large an
d complicated\, and one searches for simple limiting representations (as
the network size goes to infinity). We characterise the wealths of banks i
n a large network in terms of a simple\, one dimensional distributional fi
xed point\, which we show is amenable to simulation. While such fixed poin
t representations have been well studied when the network is of a finite s
ize\, to the best of our knowledge\, our work is the first to provide a si
mplified limiting representation.\n
URL:https://www.tcs.tifr.res.in/web/events/1079
DTSTART;TZID=Asia/Kolkata:20200827T103000
DTEND;TZID=Asia/Kolkata:20200827T113000
LOCATION:Zoom link for meeting. https://zoom.us/j/97134023137?pwd=b
XJwTkFHNFQxcDkyR2JaMkVaTk5WQT09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1080
DTSTAMP:20230914T125949Z
SUMMARY:Directed Information and Pearl's Causal Calculus
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nThis will be a tweaked ver
sion of my Qualifier talk. I will mostly focus on:\n1. The notion of inter
ventional distributions (as defined by Judea Pearl) and how they can be us
ed to identify causal linkages.\n2. How directed information can serve as
a measure of causality.\n3. How conditional directed information can be us
ed to develop an information-theoretic version of Pearl's backdoor criteri
on for identifying causality from passive observations.\n\nLink to the pap
er: https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6120270\n
URL:https://www.tcs.tifr.res.in/web/events/1080
DTSTART;TZID=Asia/Kolkata:20200828T171500
DTEND;TZID=Asia/Kolkata:20200828T181500
LOCATION:Zoom link: https://zoom.us/j/92834710011?pwd=WURSa3JsRVpJYWhONEExY
TdDcENDQT09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1081
DTSTAMP:20230914T125949Z
SUMMARY:Minimizing the Sum of Age of Information and Transmission Cost unde
r Stochastic Arrival Model
DESCRIPTION:Speaker: Kumar Saurav\n\nAbstract: \nWe consider a node-monitor
pair\, where updates are generated stochastically (according to a known d
istribution) at the node that it wishes to send to the monitor. The node i
s assumed to incur a fixed cost for each transmission\, and the objective
of the node is to find the update instants so as to minimize a linear comb
ination of AoI of information and average transmission cost. First\, we co
nsider the Poisson arrivals case\, where updates have an exponential inter
-arrival time for which we derive an explicit optimal online policy. Next\
, for arbitrary distributions of inter-arrival time of updates\, we propos
e a simple randomized algorithm that transmits any newly arrived update wi
th a fixed probability (that depends on the distribution) or never transmi
ts that update. The competitive ratio of the proposed algorithm is shown t
o be a function of the variance and the mean of the inter-arrival time dis
tribution. For some of the commonly considered distributions such as expon
ential\, uniform\, and Rayleigh\, the competitive ratio bound is shown to
be 2.\n
URL:https://www.tcs.tifr.res.in/web/events/1081
DTSTART;TZID=Asia/Kolkata:20200904T171500
DTEND;TZID=Asia/Kolkata:20200904T181500
LOCATION:Zoom link: https://zoom.us/j/92834710011?pwd=WURSa3JsRVpJYWhONEE
xYTdDcENDQT09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1082
DTSTAMP:20230914T125949Z
SUMMARY:An algebraic algorithm for minimizing linearly representable submod
ular functions
DESCRIPTION:Speaker: Rohit Gurjar (Indian Institute of Technology Bombay\nM
umbai)\n\nAbstract: \nA set function f on the subsets of a set E is called
submodular if it satisfies a natural diminishing returns property: for an
y two subsets S \\subseteq T \\subseteq E and an element x outside T\, we
have f(T + x) - f(T) \\leq f(S+x) - f(S). \nSubmodular minimization proble
m asks for finding the minimum value a given submodular function takes. Th
ere are many combinatorial optimization problems that reduce to submodular
minimization\, for example\, bipartite matching\, max flow min cut\, arbo
rescences\, disjoint spanning trees etc. All these problems admit algebrai
c algorithms\, i.e.\, algorithms involving matrix rank computations. One c
an ask what is the most general class of submodular functions whose minimi
zation admits an algebraic algorithm. Towards this\, we define the class o
f 'linearly representable submodular functions (LRSF)'\, which is based on
the rank functions of families of subspaces and captures all the above co
mbinatorial examples. We give an algebraic algorithm for minimizing LRSF.
\n\nOur algebraic algorithm for this class of functions can be parallelize
d\, and thus\, puts the problem of finding the minimizing set in the compl
exity class randomized NC. Further\, we derandomize our algorithm so that
it needs only poly-logarithmic random bits. We also identify combinatoria
l problems which are captured by LRSF minimization\, but any explicit alge
braic algorithms were not given before.\n\nJoint work with Rajat Rathi.\n
URL:https://www.tcs.tifr.res.in/web/events/1082
DTSTART;TZID=Asia/Kolkata:20200915T144500
DTEND;TZID=Asia/Kolkata:20200915T154500
LOCATION:https://zoom.us/j/93694032340?pwd=WFh1dFFlZzhPUVlQelU5c3A2UUdTQT09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1083
DTSTAMP:20230914T125949Z
SUMMARY:Parity Games are solvable in quasipolynomial time
DESCRIPTION:Speaker: Ashutosh Shankar\n\nAbstract: \nA longstanding open pr
oblem in computer science has been to show that parity games can be solved
in polynomial time. In this talk\, I will be presenting a quasipolynomial
time algorithm due to Parys (2019) which is a modification of Zielonka's
recursive algorithm (1998).\n
URL:https://www.tcs.tifr.res.in/web/events/1083
DTSTART;TZID=Asia/Kolkata:20200918T171500
DTEND;TZID=Asia/Kolkata:20200918T181500
LOCATION:https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1084
DTSTAMP:20230914T125949Z
SUMMARY:Mathematics of Neural Nets
DESCRIPTION:Speaker: Anirbit Mukherjee (Wharton School of the University of
Pennsylvania\nPennsylvania\, United States.)\n\nAbstract: \nOne of the pa
ramount mathematical mysteries of our times is to be able to explain the p
henomenon of deep-learning i.e training neural nets. Neural nets can be ma
de to paint while imitating classical art styles or play chess better than
any machine or human ever and they seem to be the closest we have ever co
me to achieving "artificial intelligence". But trying to reason about thes
e successes quickly lands us into a plethora of extremely challenging math
ematical questions - typically about discrete stochastic processes. Some o
f these questions remain unsolved for even the smallest neural nets! In th
is talk we will give a brief overview of the major themes of our work in t
his direction in the last few years.\n \nFirstly we will give highlights o
f some of our major depth hierarchy theorems and landscape results about
neural nets. Then we will explain how for certain nets under mild distribu
tional conditions our iterative algorithms like ``Neuro-Tron"\, which do n
ot use a gradient oracle\, can be proven to train nets in the infinity-nor
m loss - using as much time/sample complexity as expected from gradient ba
sed methods but in regimes where usual algorithms like (S)GD remain unprov
en. Our theorems include the particularly challenging regime of dealing wi
th non-realizable data while the net is of finite size. Next we will brief
ly look at our first-of-its-kind results about sufficient conditions for f
ast convergence of a standard adaptive gradient deep-learning algorithm\,
the RMSProp. \n\nIn the second half of the talk\, we will focus on the rec
ent rise of the PAC-Bayesian technology in being able to explain the low r
isk of certain over-parameterized nets on standardized tests. We will pres
ent our recent results in this domain which give bounds which empirically
supersede some of the existing theoretical benchmarks in this field and th
is we achieve via our new proofs about the key property of noise resilienc
e of nets. \n \nThis is joint work with Amitabh Basu (JHU)\, Ramchandran M
uthukumar (JHU)\, Jiayao Zhang (UPenn)\, Dan Roy (UToronto\, Vector Instit
ute)\, Pushpendre Rastogi (JHU\, Amazon)\, Soham De (DeepMind\, Google)\,
Enayat Ullah (JHU)\, Jun Yang (UToronto\, Vector Institute) and Anup Rao (
Adobe).\n \nBio: Anirbit Mukherjee finished his Ph.D. in applied mathemati
cs at the Johns Hopkins University advised by Prof. Amitabh Basu. He is no
w a post-doc at Wharton (UPenn)\, Statistics with Prof. Weijie Su. He spec
ializes in deep-learning theory and has been awarded 2 fellowships from JH
U for this research - the Walter L. Robb Fellowship and the inaugural Math
ematical Institute for Data Science Fellowship. Earlier\, he was a researc
her in Quantum Field Theory\, while doing his undergrad in physics at the
Chennai Mathematical Institute (CMI) and masters in theoretical physics at
the Tata Institute of Fundamental research (TIFR).\n
URL:https://www.tcs.tifr.res.in/web/events/1084
DTSTART;TZID=Asia/Kolkata:20200922T160000
DTEND;TZID=Asia/Kolkata:20200922T170000
LOCATION:Zoom meeting: https://zoom.us/j/97246120231?pwd=OGhsUTY4Unpyblkr
cUxHMnlvbGxmdz09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1085
DTSTAMP:20230914T125949Z
SUMMARY:Speed Scaling in Networks
DESCRIPTION:Speaker: Rahul Vaze\n\nAbstract: \nThe speed scaling problem ha
s been widely studied in the literature\, where there is a single server o
r a parallel bank of servers with variable speed. Choosing speed s\, incur
s a power cost given by a convex function P(s)\, whose integral is the tot
al energy consumed. The problem is to find the optimal service speed/rate
for servers that minimizes a linear combination of the flow time (total de
lay) and total energy. In this work\, we take the first steps towards desi
gning speed scaling algorithms for a network of servers. The network is de
scribed by a directed acyclic graph\, where there are multiple sources tha
t wish to send packets to their respective destinations. Algorithms are de
rived for both the worst case and stochastic job arrivals setting\, whose
competitive ratio depends only on the power functions and path diversity i
n the network\, but is independent of the workload\, and number of nodes o
f the network.\n
URL:https://www.tcs.tifr.res.in/web/events/1085
DTSTART;TZID=Asia/Kolkata:20200925T113000
DTEND;TZID=Asia/Kolkata:20200925T123000
LOCATION:Zoom Meeting https://zoom.us/j/7418257209?pwd=SGhITU5nYVBHRWhhbjc4
VXdrZjNYZz09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1086
DTSTAMP:20230914T125949Z
SUMMARY:The Kullback-Leibler Upper Confidence Bound (KLUCB) Algorithm for r
egret minimization in K-armed bandits
DESCRIPTION:Speaker: Anirban Bhattacharjee\n\nAbstract: \nThe K-armed bandi
t problem is a sequential decision making problem wherein one has to seque
ntially sample from a given set of K probability distributions (belonging
to a known family) informally called 'arms of the bandit'. The goal is to
minimize the total opportunity cost of not selecting the arm with the high
est expected reward\, called the regret. We shall look at the Kullback-Lei
bler Upper Confidence Bound (KLUCB) Algorithm for regret minimization in K
-armed bandits\, and see how it meets the lower bound on expected regret f
or our problem.\n
URL:https://www.tcs.tifr.res.in/web/events/1086
DTSTART;TZID=Asia/Kolkata:20200925T171500
DTEND;TZID=Asia/Kolkata:20200925T181500
LOCATION:Zoom link: https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0v
c0ZHcEt0Zz09
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1087
DTSTAMP:20230914T125950Z
SUMMARY:Hardness Characterisations and Size-Width Lower Bounds for QBF Reso
lution
DESCRIPTION:Speaker: Meena Mahajan (IMSc\, Chennai)\n\nAbstract: \nThis tal
k will start with an overview of the relatively young field of QBF proof c
omplexity\, explaining the QBF proof system QURes\, and an assessment of e
xisting lower bound techniques. In the main part of the talk\, I will desc
ribe a characterisation of QURes proof size in terms of a model in circuit
complexity called term decision lists\, yielding very direct connections
between circuit lower bounds and QURes proof size lower bounds. Joint work
with Olaf Beyersdorff\, Joshua Blinkhorn\, Tomáš Peitl.\nYouTube Live :
https://www.youtube.com/watch?v=Zn7ZzaELF6A\n
URL:https://www.tcs.tifr.res.in/web/events/1087
DTSTART;TZID=Asia/Kolkata:20200929T160000
DTEND;TZID=Asia/Kolkata:20200929T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1088
DTSTAMP:20230914T125950Z
SUMMARY:Convex Set Disjointness\, Distributed Learning of Halfspaces\, and
Linear Programming
DESCRIPTION:Speaker: Shay Moran (Technion)\n\nAbstract: \nDistributed learn
ing protocols are designed to train on distributed data without gathering
it all on a single centralized machine\, thus contributing to the efficien
cy of the system and enhancing its privacy.\nWe study a central problem in
distributed learning\, called Distributed Learning of Halfspaces:\nlet U
\\subset R^d be a known domain of size n and let h:R^d —> R be an unknow
n target affine function. A set of examples {(u\,b)} is distributed betwee
n several parties\, where u \\in U is a point and b = sign(h(u)) \\in {-1\
, +1} is its label.\nThe parties' goal is to agree\, using minimum communi
cation\, on a classifier f: U—>{-1\,+1} such that f(u)=b for every input
example (u\,b). (In practice\, the finite domain U is defined implicitly
by the representation of d-dimensional vectors which is used in the protoc
ol.)\nWe establish a (nearly) tight bound of ~θ(d*log n) on the communica
tion complexity of the problem of distributed learning of halfspaces in th
e two-party setting.\nSince this problem is closely related to the Convex
Set Disjointness problem in communication complexity and the problem of Di
stributed Linear Programming in distributed optimization\, we are able to
derive upper and lower bounds of ~O(d^2\\log n) and ~Ω(d\\log n) for both
of these basic problems as well.\nOur main technical contribution lies in
the design and analysis of the protocols which imply the upper bounds.\nT
o this end\, we introduce a technique called Halfspace Containers\, allowi
ng for a compressed approximate representation of every halfspace.\nHalfsp
ace containers may be of independent interest and are closely related to b
racketing numbers in statistics and to hyperplane cuttings in discrete geo
metry.\n
URL:https://www.tcs.tifr.res.in/web/events/1088
DTSTART;TZID=Asia/Kolkata:20201006T160000
DTEND;TZID=Asia/Kolkata:20201006T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1089
DTSTAMP:20230914T125950Z
SUMMARY:Diophantine Sets and Computability
DESCRIPTION:Speaker: Prabhat Kumar Jha\n\nAbstract: \nDiophantine sets are
defined using Diophantine equations. These sets are important and ubiquito
us in number theory.\nRecursively enumerable sets are defined using models
of computation (Lambda calculus\, Turing machines\, etc). These sets are
studied in computability/recursion theory.\nIn this talk\, we will see an
equivalance between these two kind of sets emerging from two different the
ories. This result is due to Matiyasevich and is known as MRDP theorem.\nT
here are many interesting corollaries of this result including Godel's inc
ompleteness theorem and undecidability of Diophantine equations.\nZoom lin
k: https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09.\n
URL:https://www.tcs.tifr.res.in/web/events/1089
DTSTART;TZID=Asia/Kolkata:20201009T171500
DTEND;TZID=Asia/Kolkata:20201009T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1090
DTSTAMP:20230914T125950Z
SUMMARY:Network Oblivious Transfer
DESCRIPTION:Speaker: Varun Narayanan\n\nAbstract: \nSecure message transmis
sion and Byzantine agreement have been studied extensively in incomplete n
etworks. However\, information theoretically secure multiparty computation
(MPC) in incomplete networks is less well understood. In this paper\, we
characterize the conditions under which a pair of parties can compute obli
vious transfer (OT) information theoretically securely against a general a
dversary structure in an incomplete network of reliable\, private channels
. We provide characterization for semi-honest model of corruption. A conse
quence of our results is a complete characterization of networks in which
a given subset of parties can compute any functionality securely with resp
ect to an adversary structure in the semi-honest case.\n
URL:https://www.tcs.tifr.res.in/web/events/1090
DTSTART;TZID=Asia/Kolkata:20201013T140000
DTEND;TZID=Asia/Kolkata:20201013T153000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1091
DTSTAMP:20230914T125950Z
SUMMARY:Some Derandomized Constructions in Quantum Information Processing
DESCRIPTION:Speaker: Aditya Nema\n\nAbstract: \nRandomness is a vital tool
and resource in both classical and quantum information processing. However
\, constructing random bits is expensive\, and hence minimising their use
is desirable. Several tools have been developed in classical computing for
this purpose. Two such tools are expanders and k-wise independent random
variables\, and they have already found many applications in classical com
puting. In this thesis we investigate the quantum analogues of these two t
ools for quantum information processing tasks.\nIn quantum information pro
cessing\, algorithms are often modeled by unitary operators and randomness
enters the picture from the choice of random unitary operators\, much lik
e randomness in classical information processing. Existential proofs in in
formation processing applications are almost always based on unitary matri
ces drawn from the Haar probability measure on the unitary group (formalis
ation of the uniform measure on the unitary group). However\, an n-qubit u
nitary is described by 4^n real parameters\, and thus cannot be efficientl
y approximated using a sub-exponential amount of time or randomness. This
has led to the construction of efficient pseudo-random ensembles\nof unita
ries which emulate the Haar measure\, up to the first t-moments\, for cert
ain applications [Harrow\, Low 2009]. Such constructions are referred to a
s unitary t-designs. It is known that there are exact unitary 't'-designs
for any t [Kuperberg\, 2006]. The advantage of using these designs is that
approximate unitary t-designs can be efficiently constructed for t = poly
(log dimension) [Brandao et al. 2012\, Sen 2018].\nIn this thesis we apply
these designs for the following tasks:\n1. Showing that the minimum outpu
t Renyi p-entropy\, for all p >= 1\, of quantum channels is subadditive. R
enyi 1-entropy is the well known von Neumann entropy and its subadditivity
leads to quantum channels with superadditive classical Holevo capacity [S
hor 2004]. We prove for the first time that approximate unitary n^{2/3}-de
signs lead to a quantum channel with superadditive classical Holevo capaci
ty\;\n2. Efficient estimation of average gate fidelity of a quantum logic
gate using few random bits\, better than the state of the art\;\n\n3. Conc
entration of measure for non-catalytic decoupling theorem in quantum infor
mation theory\, using an approximate t-design for a suitable value of t\,
beating the state of the art for relative thermalisation under certain rea
sonable assumptions.\nAlthough the unitary t-designs used in the above wor
ks have t larger than what is known to be efficiently implementable\, yet
in all cases they lead to significant savings in the number of random bits
required vis-a-vis Haar random unitaries. Thus\, all the three works can
be thought of as steps towards their respective derandomisations.\n-------
------------\nZoom Meeting https://zoom.us/j/91958942801?pwd=QzdyN2RXbE1RQ
ko1bVpMZGluYVdNUT09 Meeting ID: 919 5894 2801 Passcode: 631121\n
URL:https://www.tcs.tifr.res.in/web/events/1091
DTSTART;TZID=Asia/Kolkata:20201016T133000
DTEND;TZID=Asia/Kolkata:20201016T143000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1092
DTSTAMP:20230914T125950Z
SUMMARY:Partial Function Extension with Applications to Learning and Proper
ty Testing
DESCRIPTION:Speaker: Gunjan Kumar\n\nAbstract: \nPartial function extension
is a basic problem that underpins multiple research topics in optimizatio
n\, including learning\, property testing\, and game theory. Here\, we are
given a partial function consisting of $n$ points from a domain and a fun
ction value at each point. Our objective is to determine if this partial f
unction can be extended to a function defined on the domain\, that additio
nally satisfies a given property\, such as linearity. We formally study pa
rtial function extension for various complement-free functions.\n\nOur con
tributions are twofold. Firstly\, for the properties studied\, we give a c
ombinatorial characterization and bounds on the complexity of partial func
tion extension. Secondly\, we develop new connections between partial fun
ction extension and learning and property testing\, and use these to give
new results for these problems.\n\nZoom link: https://zoom.us/j/915860682
03?pwd=Sjk1T1lhR1lYVUhidHN2VVRXeFJoZz09\n
URL:https://www.tcs.tifr.res.in/web/events/1092
DTSTART;TZID=Asia/Kolkata:20201020T140000
DTEND;TZID=Asia/Kolkata:20201020T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1093
DTSTAMP:20230914T125950Z
SUMMARY:Learning arithmetic circuits in the average case via lower bounds
DESCRIPTION:Speaker: Ankit Garg (Microsoft Research\, Bangalore)\n\nAbstrac
t: \nThe problem of learning arithmetic circuits is the following: given a
polynomial as a black box that is promised to have a small arithmetic cir
cuit computing it\, can we find this arithmetic circuit? This problem is h
ard in the worst case and so previous works have focused on regimes where
the NP-hardness doesn't kick in (e.g. constant top fan-in etc.). We focus
on the average case where one assumes certain non-degeneracy assumptions o
n the circuit (formally these amount to assuming the circuit parameters li
e outside a certain variety and hence if they are chosen randomly accordin
g to any reasonable distribution\, the non-degeneracy condition is satisfi
ed whp). Kayal and Saha gave a meta framework (in a rudimentary form) or d
esigning these algorithms for circuit classes where we have lower bounds.
They carried this out for depth 3 circuits (sums of products of linear for
ms) and we (in joint work with Kayal and Saha) streamline their meta frame
work and carry it out for depth 4 powering circuits (sums of powers of low
degree polynomials). I will talk about the meta framework and then about
our specific results. I will also talk about a potential application to le
arning mixtures of general Gaussians.\nYouTube live: https://www.youtube.c
om/watch?v=qP1M8jmXnzc\n
URL:https://www.tcs.tifr.res.in/web/events/1093
DTSTART;TZID=Asia/Kolkata:20201020T160000
DTEND;TZID=Asia/Kolkata:20201020T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1094
DTSTAMP:20230914T125950Z
SUMMARY:A Provably Convergent and Practical Algorithm for Min-max Optimizat
ion with Applications to GANs
DESCRIPTION:Speaker: Sushant Vijayan\n\nAbstract: \nAlgorithms for finding
min-max equilibrium points in a zero sum game often exhibit cyclic behavio
r and non-convergence. The authors define a new solution concept for the z
ero sum game played through stochastic gradient descent (and similar suc
h methods) and propose an algorithm which converges with very high probabi
lity. The equilibrium point is reached in time polynomial in dimension and
smoothness parameters of the function. Importantly no convexity or concav
ity in either players control is assumed and the convergence is independen
t of the initialization point.\n\nZoom link:https://zoom.us/j/98132227553?
pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1094
DTSTART;TZID=Asia/Kolkata:20201023T171500
DTEND;TZID=Asia/Kolkata:20201023T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1095
DTSTAMP:20230914T125950Z
SUMMARY:Forecasting and the complexity of randomized algorithms
DESCRIPTION:Speaker: Eric Blais (University of Waterloo)\n\nAbstract: \nRan
domness is a remarkably powerful tool in the design of algorithms. By givi
ng algorithms the ability to use random bits and letting them err with som
e small probability\, we can solve many computational problems with rema
rkable efficiency. However\, even the most basic questions about randomize
d algorithms have proven to be remarkably challenging\, and many of those
questions even remain open to this day.\nIn this talk\, we will explore th
e notion of forecasting algorithms\, a slight variant on the usual model o
f randomized algorithms where the algorithm must generate a confidence sco
re along with its output. We will see how forecasting algorithms allow us
to bypass some of the inherent difficulties in the analysis of bounded-err
or randomized algorithms\, and lead to new developments on some of fundame
ntal problems related to the composition conjecture for query complexity a
nd to the minimax lemma.\nThis talk will cover material from joint work wi
th Shalev Ben-David that can be found at https://arxiv.org/abs/2002.10802
and https://arxiv.org/abs/2002.10809.\n
URL:https://www.tcs.tifr.res.in/web/events/1095
DTSTART;TZID=Asia/Kolkata:20201027T160000
DTEND;TZID=Asia/Kolkata:20201027T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1096
DTSTAMP:20230914T125950Z
SUMMARY:Math and Informatics\, ever more fruitful interactions
DESCRIPTION:Speaker: Gérard Berry (French Académie des sciences)\n\nAbstr
act: \nThe relation between math and algorithmics is very old. In a sense\
, one can even argue that mathematics was created to show that some alread
y known algorithms worked in all cases and not only on some examples. Chur
ch and Turing\, creators of the lambda-calculus and computability theory w
ere mathematicians. Informatics was really born later with the introductio
n of computers\, for which mathematicians played an important role. In the
70's\, computer algebra was developed\, as well as the formal semantics o
f programming languages and automatic program analysis techniques. But the
cooperation between both sciences has deepened quickly in the last 20 yea
rs\, with mutual enrichment. We will show this with a variety of examples
: the systematic development of randomized and probabilistic algorithms to
handle massive data\, the use of boolean satisfiability (SAT) systems to
solve open problems in number theory\, the development of powerful proof a
ssistants for mechanizing large mathematical proofs using rich logics (Fei
t-Thompson theorem\, Kepler conjectures)\, and finally\, the use of the sa
me tools to perform mathematical proofs of the correctness of safety-criti
cal computerized systems. Bio: Gérard Berry\, member of the French Acadé
mie des sciences and Gold Medal of CNRS in 2014\, studied at Ecole Polytec
hnique and Corps des Mines in Paris. He was researcher at Ecole des Mines
and Inria from 1970 to 2001\, Chief Scientist of the Esterel Technologies
company from 2001 to 2009\, Director of Research at Inria from 2009 to 201
2\, before being appointed as Full Professor at Collège de France in 2012
on the Algorithms\, Programs and Machine Chair\; he had previously held t
here two yearly chairs in 2007-2008 and 2009-2010. He officially retired i
n 2001. His main scientific contributions concern the formal development o
f programming languages in relation with mathematical logic\, parallel and
real-time programming\, high-level design of computer circuits and system
s\, and formal verification of programs and circuits. He is the creator of
the Esterel synchronous reactive language. Through his courses\, books an
d conferences\, he is also active in the dissemination to a general audien
ce of the new algorithmic way of thinking and acting. This virtual session
is in the framework of the Itinerant Chair in India created by the French
Institute in India ( http://ifindia.in/all-events/ ) in collaboration wit
h Collège de France ( https://www.college-de-france.fr/site/college/index
.htm ). In the heart of Paris\, Collège de France is a highly reputed Fre
nch higher education and research establishment. Since the 16th century\,
it has been a forum for cutting-edge research and education in all fields
of knowledge\, from mathematical and natural sciences to social sciences a
nd humanities. To know more about the Itinerant Chair in India : https://i
findia.in/college-de-france/\nYouTube Live : https://youtu.be/cIfE0pSWU0E
\n
URL:https://www.tcs.tifr.res.in/web/events/1096
DTSTART;TZID=Asia/Kolkata:20201103T160000
DTEND;TZID=Asia/Kolkata:20201103T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1097
DTSTAMP:20230914T125950Z
SUMMARY:Unravelling Dataset Biases
DESCRIPTION:Speaker: Mohit Lamba (Indian Institute of Technology\, Madras\n
Chennai)\n\nAbstract: \nAs scientists and engineers\, we have long aimed a
t solving real-time problems such as detecting and localizing objects seen
by our video recorder or at least something like Pokemon's Animedex. But
in recent times\, this has translated to publishing slapdash papers wherei
n the only gold standard is to beat some metric defined for a selected pro
blem as seen in the Imagenet and Pascal VOC challenge. But how good the so
-called incredible performance in such contrived situations generalizes in
a real-world setting? In this short talk\, we shall discuss several data
set biases that occur in collecting a dataset aiming to emulate the real w
orld and its consequence in narrowing down the research community's focus.
Though we shall take specific case studies from the Computer Vision commu
nity\, the hope is that the message will be equally relevant to all the di
sciples and generate curiosity to question if we have lost our original pu
rpose in a bid to break the previous benchmarks.\n\nZoom Link: https://zoo
m.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1097
DTSTART;TZID=Asia/Kolkata:20201106T171500
DTEND;TZID=Asia/Kolkata:20201106T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1098
DTSTAMP:20230914T125950Z
SUMMARY:Fair Cake Division Under Monotone Likelihood Ratios
DESCRIPTION:Speaker: Siddharth Barman (Indian Institute of Science\nBangalo
re)\n\nAbstract: \nThe cake-cutting problem provides a model for addressin
g fair allocation of a divisible resource (metaphorically\, the cake) amon
g agents with distinct preferences. Classic results of Stromquist (1980) a
nd Su (1999) show that envy-free (fair) cake divisions are guaranteed to e
xist under mild conditions. These strong existential results essentially f
ollow from fixed-point theorems and stand without an algorithmic counterpa
rt\; Stromquist (2008) has shown that an envy-free cake division with cont
iguous pieces cannot be computed in bounded time. \n\nIn this talk I will
present a result which complements these existential (and non-constructive
) guarantees by way of developing efficient cake-cutting algorithms for a
broad class of valuations. In particular\, our algorithmic result holds wh
en the agents' valuations are induced by linear translations of any log-co
ncave function\, such as Gaussian\, exponential\, linear\, or binomial.\n\
nJoint work with Nidhi Rathi. \n\nhttps://arxiv.org/abs/2006.00481\n\nhttp
s://www.youtube.com/watch?v=uzc2ht4xPFU\n
URL:https://www.tcs.tifr.res.in/web/events/1098
DTSTART;TZID=Asia/Kolkata:20201110T160000
DTEND;TZID=Asia/Kolkata:20201110T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1099
DTSTAMP:20230914T125950Z
SUMMARY:Robust Algorithms for recovering planted structures in Semi-random
instances
DESCRIPTION:Speaker: Yash Khanna (Indian Institute of Science\, Bangalore)\
n\nAbstract: \nIn this work\, we design algorithms for two fundamentally i
mportant and classical graph problems in the planted setting. Both of thes
e problems are NP-hard and the bounds known from the algorithmic front are
either not fully understood\, or not much progress can be made because of
tight lower bounds. Thus it is natural to consider semi-random models for
these problems. These models are inspired from the seminal paper of Feige
and Killian [FK01] and have been studied in numerous follow-up works with
the latest ones by Steinhardt\, and McKenzie et al. [Ste17\, MMT20]. The
construction of our instance starts with an empty graph\, then an arbitrar
y set of vertices (S) is chosen and either a dense graph or a clique (or a
n independent set) is planted on it\, the subgraph on S x V-S is a random
graph\, while the subgraph on V-S might be a random\, arbitrary\, or some
special graph (depending on the model). Our algorithms are based on roundi
ng semi-definite programs and our primary focus is on recovering (complete
ly or partially) the planted structure (S) with high probability (over the
randomness of the input). We give algorithms that exploit the geometry of
the corresponding vectors (from the SDP) and are easy to design/analyse.
\n\nThe two problems which we study are:\n1. Densest k-Subgraph/Clique\nGi
ven an undirected graph G\, the Densest k-Subgraph problem (DkS) asks to c
ompute a subset S of V of cardinality k such that the weight of edges insi
de S is maximized. This is a fundamental NP-hard problem whose approximabi
lity\, inspite of many decades of research\, is yet to be settled. There i
s a significant gap between the best known worst-case approximation factor
of this problem [BCC+10] and the hardness of approximation for it (assumi
ng the Exponential Time Hypothesis) [Man17]. We ask what are some easier i
nstances of this problem? We propose some natural semi-random models of in
stances with a planted dense subgraph\, and study approximation algorithms
for computing the densest subgraph in them. There are many such random an
d semi-random models which have been studied in the literature [BCC+10\, A
me15\, HWX16\, BA19 etc.].\n2. Independent Set in Hypergraphs\nThe indepen
dent set problem in graphs poses the following question: given a graph\, a
nd a subset of vertices such that any two vertices of the set do not have
an edge between them. The maximization version of this problem features as
one of the Karp's original twenty-one NP-complete problems ([Kar72]\, the
clique problem instead of its complement\, the independent set problem).
The independent set problem is relatively well understood and by the famo
us result of Håstad [Hås99]\, the lower and upper bounds of this problem
are tight. Hypergraphs are a natural extension of graphs\, where each edg
e is formed across a tuple of distinct vertices. For a graph\, each tuple
has a size\, two. To the best of our knowledge\, semi-random models on hyp
ergraphs have not been studied so far. Studying classical problems like th
ese on hypergraphs is naturally of theoretical as well as practical intere
st. We study the algorithmic problems studied in McKenzie et al. [MMT20] a
nd develop algorithms for them in the case of hypergraphs.\n\nNote: This i
s joint work with Anand Louis and Rameesh Paul. Both of these e-prints wil
l be up on arXiv soon.\n\nSpeaker Bio: Yash Khanna is a third (and final)
year M.Tech (Research) student in the Theory Group of CSA\, IISc Bangalore
. His research interests include Algorithms and Complexity. He previously
graduated from BITS Pilani.\n-----------------\nZoom meeting link : https:
//zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1099
DTSTART;TZID=Asia/Kolkata:20201113T171500
DTEND;TZID=Asia/Kolkata:20201113T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1100
DTSTAMP:20230914T125950Z
SUMMARY:Hitting Sets for Some Algebraic Models - Constructions and Conseque
nces
DESCRIPTION:Speaker: Anamay Tengse\n\nAbstract: \nWe study the question of
constructing _hitting sets_ for polynomials computed by several algebraic
models. For a class of polynomials C\, hitting sets for C capture the prob
lem of _deterministic_ blackbox PIT for C - checking if a polynomial from
C is zero by querying it on a few points. Formally\, a hitting set for a c
lass C is a set H such that for every nonzero f in C\, there is some point
h in H for which f(h) is nonzero. Owing to its close connections to the s
earch of explicit hard polynomials\, finding efficient hitting sets for va
rious classes of polynomials is a central question in algebraic complexity
theory.\nOur contributions towards this question are as follows. The firs
t set of our results extend the scope of known hitting set constructions t
o other well studied algebraic models. Next\, we show that for any general
enough algebraic model\, like circuits or formulas\, even a mild improvem
ent to the trivial hitting sets for the model leads to almost efficient hi
tting sets for it. Lastly\, we explore how hitting sets for a class of pol
ynomials can assist us in _proving_ lower bounds against that class.\n
URL:https://www.tcs.tifr.res.in/web/events/1100
DTSTART;TZID=Asia/Kolkata:20201123T111500
DTEND;TZID=Asia/Kolkata:20201123T121500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1101
DTSTAMP:20230914T125950Z
SUMMARY:Optimal Control Problems with Symmetry Breaking Cost Functions
DESCRIPTION:Speaker: Rohit Gupta (University of Michigan)\n\nAbstract: \nSy
mmetry reduction is a well studied subject in geometric mechanics\, where
symmetries are usually described as an invariance under an action of a Lie
group. Symmetry breaking is also common in several physical contexts\, fr
om classical mechanics to particle physics and in certain cases\, it is st
ill possible to carry out symmetry reduction. The simplest example is the
heavy top dynamics (the motion of a rigid body with a fixed point in a gra
vitational field)\, where due to the presence of gravity\, we get a Lagran
gian that is SO(2)-invariant but not SO(3)-invariant\, contrary to what ha
ppens for the free rigid body. Based on the ideas of symmetry reduction st
udied in geometric mechanics\, symmetry reduction of optimal control probl
ems (OCPs) for left-invariant control systems on Lie groups has been studi
ed extensively over the past couple of decades and by exploiting these sym
metries\, the system can be reduced to a lower-dimensional one or decouple
d into subsystems. In this talk\, I will discuss symmetry reduction of OCP
s for left-invariant control affine systems on Lie groups with partially b
roken symmetries\, more specifically\, cost functions that break some but
not all of the symmetries. I will illustrate the theory with the motion pl
anning problem of a controlled unicycle (a popular model used in robotics)
in the presence of an obstacle\, where the symmetry breaking appears natu
rally in the form of a barrier function.\n\nBio: Rohit Gupta received the
M.S.E. degree in Mechanical Engineering and the Ph.D. degree in Aerospace
Engineering from the University of Michigan in 2012 and 2016\, respectivel
y. From January 2016 to August 2017\, he was a postdoctoral research assoc
iate at the Institute for Mathematics and its Applications (IMA) at the Un
iversity of Minnesota. From August 2017 to August 2018\, he was a postdoct
oral research associate in the department of Mathematics at Michigan State
University. From September 2018 to August 2019\, he was a postdoctoral re
search fellow in the department of mechanical engineering at the Universit
y of Michigan. From September 2019 to August 2020\, he was a postdoctoral
research fellow in the department of electrical engineering and computer s
cience at the University of Michigan. His research interests include geome
tric mechanics\, geometric control\, optimal control and optimization.\n
URL:https://www.tcs.tifr.res.in/web/events/1101
DTSTART;TZID=Asia/Kolkata:20201124T140000
DTEND;TZID=Asia/Kolkata:20201124T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1102
DTSTAMP:20230914T125950Z
SUMMARY:Degree vs Approximate Degree for Boolean Functions
DESCRIPTION:Speaker: Suhail Sherif\n\nAbstract: \nThe field of researching
Boolean functions can be summed up as\nBoolean functions: *exist*\nResearc
hers: find out EVERYTHING\nThis has led to it being a very happening field
with many elegant results and fascinating stories. The algebraic properti
es of degree and approximate degree have played prominent roles in many of
these. A couple of years before I was born\, Gotsman and Linial showed a
purely combinatorial reformulation of the `degree vs sensitivity' question
.\nLast year Hao Huang shocked the world by proving this combinatorial sta
tement with a linear algebraic argument that fits in a tweet.\nCome this y
ear and the team of Scott Aaronson\, Shalev Ben-David\, Robin Kothari\, Sh
ravas Rao and Avishay Tal have delved into this linear algebraic argument
and proved many other implications of it\, the one most interesting to me
being\napprox-degree(f) ≥ Ω(√degree(f))\nIt is a quite elementary pro
of taking advantage of intuitive linear algebraic manipulations. This is w
hat I intend to present in this student seminar.\n\nZoom Link:\nhttps://zo
om.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1102
DTSTART;TZID=Asia/Kolkata:20201127T171500
DTEND;TZID=Asia/Kolkata:20201127T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1103
DTSTAMP:20230914T125950Z
SUMMARY:SNARGs and PPAD Hardness from Sub-exponential LWE
DESCRIPTION:Speaker: Prof. Dakshita Khurana (University of Illinois at Urba
na-Champaign)\n\nAbstract: \nWe construct a succinct non-interactive publi
cly-verifiable delegation scheme for any logspace uniform circuit under th
e sub-exponential Learning With Errors (LWE) assumption. For a circuit C :
{0\, 1}^N -> {0\, 1} of size S and depth D\, the prover runs in time poly
(S)\, the communication complexity is D * polylog(S)\, and the verifier ru
ns in time (D + N) * polylog(S). To obtain this result\, we introduce a ne
w cryptographic primitive: lossy correlation-intractable hash functions. W
e use this primitive to soundly instantiate the Fiat-Shamir transform for
a large class of interactive proofs\, including the interactive sum-check
protocol and the GKR protocol\, assuming the sub-exponential hardness of L
WE. By relying on the result of Choudhuri et al. (STOC 2019)\, we also est
ablish the sub-exponential average-case hardness of PPAD\, assuming the su
b-exponential hardness of LWE. This is based on joint work with Ruta Jawal
e\, Yael Tauman Kalai and Rachel Zhang.\nYouTube Link- https://www.youtube
.com/watch?v=H7zwkYuILiQ\n
URL:https://www.tcs.tifr.res.in/web/events/1103
DTSTART;TZID=Asia/Kolkata:20201208T173000
DTEND;TZID=Asia/Kolkata:20201208T183000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1104
DTSTAMP:20230914T125950Z
SUMMARY:Optimization for scheduling and beyond\, a set covering view of sch
eduling
DESCRIPTION:Speaker: Jatin Batra (CWI Amsterdam\, Netherlands)\n\nAbstract:
\nScheduling provides an interesting context for optimization methods. Th
is talk is about how a set covering viewpoint can provide powerful insight
s into scheduling in many settings - minimizing response times on a single
machine\, heterogenous scheduling on multiple machines\, diversification
in information retrieval etc. Along the way\, I will also illustrate conne
ctions and improvements to classic set covering problems in the context of
knapsacks\, geometric set cover and minimum-cuts. Along the way\, I will
show how Chernoff bounds can be improved for lower tails of Bernoulli dist
ributions.\n
URL:https://www.tcs.tifr.res.in/web/events/1104
DTSTART;TZID=Asia/Kolkata:20201210T160000
DTEND;TZID=Asia/Kolkata:20201210T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1105
DTSTAMP:20230914T125950Z
SUMMARY:Off-policy evaluation in Reinforcement Learning using Linear Regres
sion
DESCRIPTION:Speaker: Anirban Bhattacharjee\n\nAbstract: \nIn Reinforcement
Learning\, one often needs to evaluate a given policy using rewards observ
ed by following another policy. This is called off-policy evaluation in Le
arning Theory parlance. The traditional methods for off-policy evaluation
involve importance sampling\, which comes with certain drawbacks. We shall
look at these drawbacks and how linear regression may be used instead to
overcome the same.\n\nZoom link: https://zoom.us/j/98132227553?pwd=K2cyQll
KVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1105
DTSTART;TZID=Asia/Kolkata:20201211T171500
DTEND;TZID=Asia/Kolkata:20201211T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1106
DTSTAMP:20230914T125950Z
SUMMARY:Formalizing Finite Set Combinatorics in Type Theory
DESCRIPTION:Speaker: Abhishek Singh\n\nAbstract: \nMathematical proofs when
written in conventional ways often contain imprecise definitions\, unstat
ed background assumptions\, and inferential gaps in reasoning. In such cir
cumstances\, it becomes difficult for a reviewer to determine whether the
given proof is correct or not. Even if the theorem statement turns out t
o be true\, judging it to be so could take a long time. A possible solutio
n to this problem is to formalize mathematical results using Proof Assista
nts. Proof Assistants are software tools built on top of a small and trust
ed kernel that provides a formal language for writing mathematical stateme
nts and their proofs. Hence\, formally verifying a mathematical theory usi
ng a Proof Assistant can increase our confidence in the verified results.
However\, the task of formalizing mathematics using Proof Assistants prese
nts some unique challenges\; both practical as well as theoretical. Theore
tical challenges mostly arise because some trivially assumed axioms of cla
ssical mathematics may not be provable in the core logic of the Proof Assi
stant. On the other hand\, the practical difficulties of formalization mos
tly arise because the machine-checkable proofs are significantly more deta
iled than the corresponding paper proofs. In this talk\, we will address b
oth these issues while presenting formal proofs of some key results from f
inite set combinatorics. More precisely\, we present formalized libraries
of definitions and results on two important mathematical structures from c
ombinatorics: (i) finite partially ordered sets\, and (ii) finite simple g
raphs. These libraries have been formalized in the type theory of Coq Proo
f Assistant.\n\nZoom Link:\nhttps://zoom.us/j/95365190349?pwd=ZENsRG1pbmVp
V01nQ3BPRGJVZmZPdz09\nMeeting ID: 953 6519 0349\nPasscode: 993493\n
URL:https://www.tcs.tifr.res.in/web/events/1106
DTSTART;TZID=Asia/Kolkata:20201214T103000
DTEND;TZID=Asia/Kolkata:20201214T113000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1107
DTSTAMP:20230914T125950Z
SUMMARY:Accelerating Black Box Estimation of Distribution Tails Using Self
structuring Importance Samplers
DESCRIPTION:Speaker: Anand Deo\n\nAbstract: \nMotivated by the increasing a
doption of models which facilitate greater automation in risk management a
nd decision-making\, this talk presents a novel Importance Sampling (IS) s
cheme for estimating distribution tails for a rich class of objectives mod
elled with tools such as mixed integer linear programs\, deep neural netwo
rks\, etc. A key challenge with the conventional efficient sampling approa
ches in these settings is the need to intricately tailor the samplers base
d on the underlying probability distribution and the objective. This chall
enge is overcome in the proposed black-box scheme by automating the select
ion of an effective IS density with a transformation that implicitly learn
s and replicates the concentration properties observed in less rare sample
s. Despite its simple and scalable implementation\, this self structuring
IS scheme achieves asymptotically optimal variance reduction across a spe
ctrum of multivariate distributions involving light as well as heavy tails
.\n\nThis approach is guided by a large deviations principle that brings o
ut the phenomenon of self-similarity of optimal IS distributions in consid
erable generality. In addition to helping certify variance reduction\, th
e large deviations principle serves as a tool for readily yielding new tai
l risk asymptotics and algorithms in settings such as distribution network
s.\n\nThis is joint work with Karthyek Murthy (SUTD).\n\nZoom link: https:
//zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1107
DTSTART;TZID=Asia/Kolkata:20201218T171500
DTEND;TZID=Asia/Kolkata:20201218T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1108
DTSTAMP:20230914T125950Z
SUMMARY:Algebraic Natural Proofs: the plot thickens
DESCRIPTION:Speaker: Anamay Tengse\n\nAbstract: \nIn the late 1990s\, a pap
er by Razborov and Rudich pointed out a barrier towards proving boolean ci
rcuit lower bounds. They observed that almost all of the then known lower
bounds could be proved using a specific template: the 'natural proofs fram
ework'. They further showed that any 'natural proof' for P/poly (poly-size
d boolean circuits) would contradict the existence of One-Way-Functions: a
crucial primitive in modern day cryptography. Thus\, their paper essentia
lly ruled out "natural" techniques for proving strong boolean circuit lowe
r bounds.\n\nAbout two decades later\, two groups: Grochow\, Kumar\, Saks
and Saraf (2017)\, and Forbes\, Shpilka and Volk (2018)\, independently pr
oposed the framework of 'algebraically natural proofs' for algebraic circu
its. They observed that most of the known lower bounds against algebraic c
ircuit models fit in this framework\, and showed that 'algebraically natur
al proofs' do not exist for VP (poly-sized algebraic circuits)\, under a (
non-standard) derandomisation assumption. Forbes\, Shpilka and Volk (2018)
also provided some evidence that suggested that the derandomisation assum
ption might indeed be true.\n\nThis year\, in joint works with Chatterjee\
, Kumar\, Ramya and Saptharishi\, we observed the following seemingly cont
radictory facts about algebraically natural proofs.\n1. Algebraically natu
ral proofs (kind of) *exist* for all "interesting" polynomials in VP and V
NP (algebraic P and NP).\n2. Algebraically natural proofs *do not exist* f
or VNP (algebraic NP)\, if permanent is exponentially hard.\nIn this talk\
, we will first understand the algebraically natural proofs framework\, an
d then try to make sense of our recent findings.\n\nZoom link: https://zoo
m.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1108
DTSTART;TZID=Asia/Kolkata:20210101T171500
DTEND;TZID=Asia/Kolkata:20210101T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1109
DTSTAMP:20230914T125950Z
SUMMARY:Singularly Optimal Randomized Leader Election
DESCRIPTION:Speaker: William K. Moses Jr. (University of Houston\, USA)\n\n
Abstract: \nThis paper concerns designing distributed algorithms that ar
e singularly optimal\, i.e.\, algorithms that are simultaneously time and
message optimal\, for the fundamental leader election problem in network
s. Our main result is a randomized distributed leader election algorithm f
or asynchronous complete networks that is essentially (up to a polylogarit
hmic factor) singularly optimal. Our algorithm uses O(n) messages with hig
h probability and runs in O(\\log^2 n) time (with high probability) to ele
ct a unique leader. The O(n) message complexity should be contrasted with
the \\Omega(n \\log n) lower bounds for the deterministic message complexi
ty of leader election algorithms (regardless of time)\, proven by Korach\,
Moran\, and Zaks (TCS\, 1989) for asynchronous algorithms and by Afek and
Gafni (SIAM J. Comput.\, 1991) for synchronous networks. Hence\, our resu
lt also separates the message complexities of randomized and deterministic
leader election. More importantly\, our (randomized) time complexity of O
(\\log^2 n) for obtaining the optimal O(n) message complexity is significa
ntly smaller than the long-standing \\tilde{\\Theta}(n) time complexity ob
tained by Afek and Gafni and by Singh (SIAM J. Comput.\, 1997) for messa
ge optimal (deterministic) election in asynchronous networks. Afek and Gaf
ni also conjectured that \\tilde{\\Theta}(n) time would be optimal for mes
sage-optimal asynchronous algorithms. Our result shows that randomized alg
orithms are significantly faster.\nTurning to synchronous complete network
s\, Afek and Gafni showed an essentially singularly optimal deterministic
algorithm with O(\\log n) time and O(n \\log n) messages. Ramanathan et al
. (Distrib. Comput. 2007) used randomization to improve the message comple
xity\, and showed a randomized algorithm with O(n) messages but still with
O(\\log n) time (with failure probability O(1 / \\log^{\\Omega(1)}n)). Ou
r second result shows that synchronous complete networks admit a tightly s
ingularly optimal randomized algorithm\, with O(1) time and O(n) messages
(both bounds are optimal). Moreover\, our algorithm's time bound holds wit
h certainty\, and its message bound holds with high probability\, i.e.\, 1
-1/n^c for constant c.\nOur results demonstrate that leader election can b
e solved in a simultaneously message and time-efficient manner in asynchro
nous complete networks using randomization. It is open whether this is pos
sible in asynchronous general networks.\nThis talk is based on joint work
with Shay Kutten\, Gopal Pandurangan\, and David Peleg which was published
at DISC 2020.\nBio: William K. Moses Jr. is currently a postdoc at the U
niversity of Houston in Houston\, USA\, where he is hosted by Gopal Pandur
angan. Prior to that\, he was a postdoc at the Technion and received his P
hD from IIT Madras in 2018. He is interested in the theory of distributed
computing with a focus on distributed algorithms. His three current areas
of focus are problems in general graphs\, programmable matter\, and mobile
robots. He has in the past worked on problems related to SINR networks an
d load balancing. His profile can be accessed here: https://sites.google
.com/view/wkmjr/home\nZoom link:\nhttps://zoom.us/j/98132227553?pwd=K2cyQl
lKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1109
DTSTART;TZID=Asia/Kolkata:20210109T100000
DTEND;TZID=Asia/Kolkata:20210109T110000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1110
DTSTAMP:20230914T125950Z
SUMMARY:Hardness of Recognizing Geometric Intersection Graphs
DESCRIPTION:Speaker: Kshitij Gajjar (Technion - Israel Institute of Technol
ogy\nHaifa\, Israel.)\n\nAbstract: \nMany graph problems that are NP-hard
for general graphs can be solved in polynomial time for planar graphs. We
explore the domain of "almost" planar graphs. These are graphs that can be
made planar by removing one or two vertices from them. We show that recog
nition of intersection graphs for several different types of geometric obj
ects in the plane (e.g.\, line segments\, elliptical disks\, rectangles\,
simple curves\, etc.) is NP-hard\, even if the inputs are restricted to al
most planar graphs.\n\nNo background other than high school geometry will
be assumed for this talk.\nThis is joint work with Dibyayan Chakraborty.\n
\nZoom link:\nhttps://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcE
t0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1110
DTSTART;TZID=Asia/Kolkata:20210115T171500
DTEND;TZID=Asia/Kolkata:20210115T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1111
DTSTAMP:20230914T125951Z
SUMMARY:Secure Multiparty Computation with Limited Connectivity
DESCRIPTION:Speaker: Varun Narayanan\n\nAbstract: \nInformation theoretical
ly secure multiparty computation (MPC) is a central primitive in modern cr
yptography.\nIt enables mutually distrusting parties to collaboratively pe
rform computations on their combined data by ensuring that each party's da
ta is kept private from the others.\nThis is achieved by designing communi
cation protocols which allow the parties to collectively simulate an incor
ruptible trusted party\, who privately receives inputs from the parties\,
computes the pre-agreed functionality\, and delivers the outputs to the ap
propriate parties privately.\nThe subject of this dissertation is MPC when
there is limited connectivity in the communication network available to t
he participants.\nOur motivations and the progress we made in addressing t
hem follows:\n- In many practical scenarios\, the parties may only have ac
cess to a communication network with limited connectivity\, in that\, not
every pair of parties can communicate privately and reliably with each oth
er.\nWe characterize the conditions under which a pair of parties can comp
ute any functionality with information theoretic security in an incomplete
network of reliable\, private links.\nSeparate characterizations are obta
ined for honest-but-curious and malicious modes of corruption with securit
y against general adversary structures.\n- Many cryptographic tasks can be
modelled as secure 2-party computation (2PC) using only one-directional c
ommunication.\nGarg et al. (Crypto 15) initiated the study of non-interact
ive 2PC over noisy channels with one-way communication\, namely when only
one party speaks.\nA major question left open by that work was the complet
eness of finite channels in this model of secure computation.\nWe show tha
t bit-ROT (i.e.\, Randomized Oblivious Transfer) channel\, which erases on
e of the two input bits uniformly at random\, can compute any functionalit
y with inverse polynomial security error (in the number of channel uses) i
n this model against a computationally unbounded adversary.\nFurther\, ass
uming ideal obfuscation\, realizable using tamper-proof hardware tokens\,
naturally occurring channels such as binary symmetric channel (BSC) and bi
nary erasure channel (BEC) are complete in this sense with inverse polynom
ial security error against a computationally bounded adversary.\nTo comple
ment this\, we show that no channel with finite alphabet is complete in th
is model with negligible security error even against a computationally bou
nded adversary.\nFinally\, we characterize the channels that enable zero-k
nowledge proofs in this model\; the previous result work had presented con
struction of zero-knowledge proofs using BEC/BSC channels.\n- Studying sec
ure computation with limited interaction tends to reveal new frontiers to
approach the problem of complexity of several information theoretic primit
ives: a notoriously hard problem in cryptography.\nWe introduce a new prim
itive in information-theoretic cryptography\, namely zero-communication re
ductions (ZCR)\, with varying levels of security\, and relate it with seve
ral other important primitives.\nUsing these connections\, we obtain new u
pper bounds and lower bounds for complexity of several cryptographic primi
tives.\n- MPC provides a meaningful and robust definition of security that
can be used for modelling security guarantees for existing models in netw
ork information theory.\nIndex coding is a well studied problem\, in which
a server wants to efficiently broadcast n messages intended for n users\,
each with access to a subset of these messages as side information.\nWe i
ntroduce a notion of privacy in index coding\, where the receivers do not
learn anything more than the message they want from the server and those t
hey have as side information\, and study various aspects of its transmissi
on rate and secret consumption rate.\n
URL:https://www.tcs.tifr.res.in/web/events/1111
DTSTART;TZID=Asia/Kolkata:20210118T110000
DTEND;TZID=Asia/Kolkata:20210118T120000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1112
DTSTAMP:20230914T125951Z
SUMMARY:Approximating a polynomial as a sum of simple polynomials
DESCRIPTION:Speaker: Neeraj Kayal (Microsoft Research)\n\nAbstract: \nIn th
is talk\, we will consider algorithmic problems which follow the following
template: given a real-valued multivariate polynomial f(x) of degree d\,
is it approximately equal to a sum of a few "simple" polynomials\, i.e Is
f ~= g_1(x) + g_2(x) + ... + g_r(x)? Examples/special cases of this proble
m template are low-rank approximation of a matrix and tensor decomposition
. We will see many applications including independent component analysis\,
subspace clustering\, Learning Gaussian mixture models and (language) top
ic modelling. In the next part\, we will see how techniques from algebraic
complexity can potentially be used to algorithmically solve such problems
efficiently. We will formulate some conjectures in this regard. We resolv
e one such conjecture which leads to a more noise-resilient algorithm for
the relatively well-studied problem of tensor decomposition.\nYouTube Link
- https://www.youtube.com/watch?v=ZCU74TXmG9o\n
URL:https://www.tcs.tifr.res.in/web/events/1112
DTSTART;TZID=Asia/Kolkata:20210126T160000
DTEND;TZID=Asia/Kolkata:20210126T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1113
DTSTAMP:20230914T125951Z
SUMMARY:An Introduction to Chordal Graphs And Clique Trees
DESCRIPTION:Speaker: Vidya Sagar Sharma\n\nAbstract: \nAn undirected graph
is chordal if every cycle of length greater than three has a chord: namely
\, an edge connecting two nonconsecutive vertices on the cycle. A clique
of a graph $G$ is any maximal set of vertices that is complete in $G$.
Let $G$ be a chordal graph and $K_G = \\{ K_1\, K_2\, ...\, K_m \\}$ den
otes the set containing the cliques of $G$\, then a tree with vertex set $
K_G$ is said to be a clique tree of the chordal graph $G$ if it follows th
e clique intersection property: For every pair of distinct cliques $K\,K'
\\in K_G$\, the set $K \\cap K'$ is contained in every clique on the path
connecting $K$ and $K'$ in the tree.\n\nIn this talk\, we will see a few p
roperties of the chordal graphs and the clique trees of the chordal graphs
.\nReference: link.springer.com/content/pdf/10.1007%2F978-1-4613-8369-7_1.
pdf\n\nZoom Link : https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc
0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1113
DTSTART;TZID=Asia/Kolkata:20210130T171500
DTEND;TZID=Asia/Kolkata:20210130T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1114
DTSTAMP:20230914T125951Z
SUMMARY:Differential Games in Spread of Diseases.
DESCRIPTION:Speaker: Sushant Vijayan\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1114
DTSTART;TZID=Asia/Kolkata:20210205T093000
DTEND;TZID=Asia/Kolkata:20210205T103000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1115
DTSTAMP:20230914T125951Z
SUMMARY:An Exploration of R ́enyi Divergence Based Upper Bounds on General
ization Error
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1115
DTSTART;TZID=Asia/Kolkata:20210205T103000
DTEND;TZID=Asia/Kolkata:20210205T113000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1116
DTSTAMP:20230914T125951Z
SUMMARY:Paper: Pseudorandom Generators from Polarizing Random Walks by Esha
n Chattopadhyay\, Pooya Hatami\, Kaave Hosseini\, and Shachar Lovett
DESCRIPTION:Speaker: Ashutosh Shankar\n\nAbstract: \nhttps://theoryofcomput
ing.org/articles/v015a010/\n
URL:https://www.tcs.tifr.res.in/web/events/1116
DTSTART;TZID=Asia/Kolkata:20210205T113000
DTEND;TZID=Asia/Kolkata:20210205T123000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1117
DTSTAMP:20230914T125951Z
SUMMARY:Element distinctness using arbitrary binary gates
DESCRIPTION:Speaker: Siddharth Bhandari\n\nAbstract: \nWe will study the De
cision-Tree complexity of element distinctness using arbitrary binary gate
s (an instance of which is comparison gates). Concretely\, let $m$ and $n$
be natural numbers with $m>n$. Suppose\, we are given an array $A[1]\,A[2
]\,\\ldots\,A[n]$ where each $A[i]\\in [m]$. At each step we are allowed t
o pick two indices $i$ and $j$ in $[n]$ and an arbitrary function $f:\\N \
\times \\N \\to \\{0\,1\\}$\, and ask for the value of $f(A[i]\,A[j])$. At
the end of the procedure we must be able to tell whether all the entries
of $A$ were distinct or not. Define by $S(m\,n)$ the minimum number of ste
ps needed in such a procedure. We are interested in understanding the asym
ptotics of $S(m\,n)$.\nIn the specific case when we are restricted to use
only comparison gates\, i.e.\, $f(a\,b)=[a>b]$ we will see that $S(m\,n)=\
\Theta(n \\log n)$ for all $m\\geq n$.\nIn the more general setting where
we are allowed to use arbitrary gates\, we will see that when $m>>n$ then
$S(m\,n)=\\Theta(n \\log n)$. However\, when $m=n$ our understanding is in
complete: $S(m\,n)=\\Omega(n \\sqrt(\\log n))$. (This is a result of Ravi
Boppana https://www.sciencedirect.com/science/article/pii/0020019094001545
?via%3Dihub). If time permits we might discuss a few ideas to bridge this
gap.\nZoom link:\nhttps://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0
ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1117
DTSTART;TZID=Asia/Kolkata:20210212T171500
DTEND;TZID=Asia/Kolkata:20210212T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1118
DTSTAMP:20230914T125951Z
SUMMARY:Banach-Tarski Paradox
DESCRIPTION:Speaker: Abhishek Khetan\n\nAbstract: \nIn this talk we will gi
ve a proof of the fact that the two dimensional sphere can be partitioned
into finitely many pieces in such a way that a rearrangement of the pieces
produces two disjoint copies of the original sphere.\nZoom link:\nhttps:/
/zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1118
DTSTART;TZID=Asia/Kolkata:20210219T171500
DTEND;TZID=Asia/Kolkata:20210219T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1119
DTSTAMP:20230914T125951Z
SUMMARY:Existence of Strategy in Games
DESCRIPTION:Speaker: Prabhat Kumar Jha\n\nAbstract: \nGames are used to mod
el many instances arising from interaction of more than one computational
agent. In program synthesis\, existence of strategy is the key in deciding
the existence of a program with a given set of specifications.\n\nIn this
talk\, we will observe Martin's proof of the existence of strategy in a c
lass of two-player infinite games with perfect information (namely Borel g
ames). One instance of this result is that in the initial stage of a game
of chess\, at least one of the players has a strategy to prevent from losi
ng.\n\nLink of the paper: https://www.jstor.org/stable/pdf/1971035.pdf?ref
reqid=excelsior%3Aed93a87cd71602b5a789b6280bf1a5a6\n
URL:https://www.tcs.tifr.res.in/web/events/1119
DTSTART;TZID=Asia/Kolkata:20210226T150000
DTEND;TZID=Asia/Kolkata:20210226T160000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1120
DTSTAMP:20230914T125951Z
SUMMARY:Tensor Rank and Algebraic Formula Lower Bounds
DESCRIPTION:Speaker: Prerona Chatterjee\n\nAbstract: \nTensor are higher d
imensional analogues of matrices and there is a notion of the rank of a te
nsor (similar to matrices). However\, unlike matrices\, finding the rank o
f a tensor is NP-hard and it is a big open problem to find explicit tensor
of large rank. Raz showed that in certain regimes\, finding such an expli
cit tensor would yield lower bounds against algebraic formulas (a natural
model for computing polynomials). In this talk\, we will introduce tensors
and view them as some structured polynomials. We will then go over the pr
oof idea of the statement mentioned above and completely prove one of the
main structural lemmas used. The talk is based on the presentation of Raz'
s paper (https://dl.acm.org/doi/10.1145/2535928) as given in chapter 16 of
Ramprasad's survey (https://github.com/dasarpmar/lowerbounds-survey/relea
ses/download/v8.0.7/fancymain.pdf). Zoom link: https://zoom.us/j/981322275
53?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1120
DTSTART;TZID=Asia/Kolkata:20210312T171500
DTEND;TZID=Asia/Kolkata:20210312T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1121
DTSTAMP:20230914T125951Z
SUMMARY:STCS Annual Symposium
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1121
DTSTART;TZID=Asia/Kolkata:20210319T090000
DTEND;TZID=Asia/Kolkata:20210320T180000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1122
DTSTAMP:20230914T125951Z
SUMMARY:Chasing Convex Functions
DESCRIPTION:Speaker: Anupam Gupta (Carnegie Mellon University)\n\nAbstract:
\nThe problem of chasing convex functions is easy to state: faced with a
sequence of convex functions {f_t}\, the goal of the algorithm is to outpu
t a point x_t at each time\, so that the sum of the function costs f_t(x_t
)\, plus the movement costs || x_t - x_{t-1} || is minimized. This general
problem generalizes several classic questions in online algorithms\, such
as caching and the k-server problem.\nThe question of getting an algorith
m whose total cost is comparable to the optimal cost in hindsight was pose
d by Friedman and Linial in 1994. This was finally (mostly) resolved last
year\, using a combination of ideas from online algorithms and convex geom
etry. In this talk we will survey the results and ideas.\nThis is based on
works with C.J.Argue\, S.Bubeck\, M.B.Cohen\, G.Guruganesh\, Y. T.Lee\, a
nd Z.Tang.\nYouTube link : https://www.youtube.com/watch?v=TMilw8CX1Yg\n
URL:https://www.tcs.tifr.res.in/web/events/1122
DTSTART;TZID=Asia/Kolkata:20210323T160000
DTEND;TZID=Asia/Kolkata:20210323T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1123
DTSTAMP:20230914T125951Z
SUMMARY:Generalization and Guessing
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nAbstract: Generalization e
rror is the gap between an algorithm's performance on the true data distri
bution (unknown to us) and its performance on the given dataset (known to
us). Thus\, establishing upper bounds on generalization error is naturally
of interest. In 2020\, Steinke and Zakynthinou derived their bound in ter
ms of the ability to guess an algorithm's input by observing its output. I
n this talk\, we will try to go over the proof of this result.\nLink to th
e paper: https://arxiv.org/abs/2001.09122\n\nZoom link:\nhttps://zoom.us/j
/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1123
DTSTART;TZID=Asia/Kolkata:20210326T171500
DTEND;TZID=Asia/Kolkata:20210326T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1124
DTSTAMP:20230914T125951Z
SUMMARY:Two-and-a-half proofs of "the duality trick"
DESCRIPTION:Speaker: Anamay Tengse\n\nAbstract: \nThe fact that the polynom
ial (x1+...+xn)^d can be written as a poly(n\,d)-sum of products of univar
iates is a consequence of what is popularly known as 'the duality trick' i
n the algebraic complexity circles. The actual identity is much more gener
al\, and in particular\, applies to any linear form raised to an arbitrary
power. Saxena (2008) famously used this duality to derive identity testin
g algorithms for polynomials that are poly(n\,d)-sums of powers of linear
forms. This model (AKA sum-power-sum) coincides with the well-studied noti
on of Waring rank in mathematics.\n\nIn this talk\, we will see two proofs
of the duality trick due to Saxena and Shpilka respectively. Both these p
roofs use fairly elementary ideas about polynomials. We shall then see a d
ifferent proof of correctness of Saxena's construction that uses mildly no
n-elementary algebraic geometry.\n\nZoom link: https://zoom.us/j/981322275
53?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1124
DTSTART;TZID=Asia/Kolkata:20210409T171500
DTEND;TZID=Asia/Kolkata:20210409T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1125
DTSTAMP:20230914T125951Z
SUMMARY:Network Coding Conjecture implies Data Structure Lower Bounds
DESCRIPTION:Speaker: Pavel Dvorak (Charles University\, Prague)\n\nAbstract
: \nNetwork coding conjecture (NCC) by Li and Li asserts that network codi
ng for undirected graphs does not bring any advantage over multicommodity
flows. Recently\, NCC was used to prove a conditional lower bound for sort
ing algorithms with external memory [Farhadi et al.\, STOC 2019]\, circuit
s for multiplication [Afshani et al.\, ICALP 2019]\, and circuits for sort
ing [Asharov et al.\, SODA 2021]. We use a technique of Farhadi et al. to
prove an NCC-based lower bound for non-adaptive data structures for functi
on inversion\, polynomial evaluation\, and polynomial interpolation. This
is a joint work with Michal Koucký\, Karel Král and Veronika Slívová.\
n
URL:https://www.tcs.tifr.res.in/web/events/1125
DTSTART;TZID=Asia/Kolkata:20210415T174500
DTEND;TZID=Asia/Kolkata:20210415T184500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1126
DTSTAMP:20230914T125951Z
SUMMARY:Fair Cake Division
DESCRIPTION:Speaker: Nidhi Rathi (IISc Bangalore)\n\nAbstract: \nThe theory
of Fair Division addresses the fundamental problem of allocating goods am
ong agents with equal entitlements but distinct preferences. The classic c
ake-cutting problem provides a model for addressing fair and efficient all
ocation of a divisible\, heterogeneous resource (metaphorically\, the cake
) among agents with varied preferences. Classic results of Stromquist (198
0) and Su (1999) show that envy-free (fair) cake divisions (with contiguou
s pieces) are guaranteed to exist under mild conditions. These strong exis
tential results follow from fixed-point theorems and stand without an algo
rithmic counterpart.\n\nIn this talk\, I will present two of the recent re
sults that complements the existential (and non-constructive) guarantees a
nd various hardness results either by developing polynomial-time approxima
tion algorithms or by identifying computationally tractable instances for
fair cake division. Our work identifies a broad class of cake division ins
tances that essentially admits a polynomial time algorithm for computing f
air and efficient allocations. In particular\, our algorithmic result hold
s when (all) agents' valuations are induced either by linear translations
of any log-concave function\, Gaussian\, exponential\, linear\, or binomia
l distributions.\n\nJoint work with Siddharth Barman\, Eshwar Ram Arunchal
eswaran and Rachitesh Kumar.\n\nhttps://arxiv.org/abs/2006.00481\nhttps://
arxiv.org/abs/1907.11019\n\nZoom link: https://zoom.us/j/98132227553?pwd=K
2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1126
DTSTART;TZID=Asia/Kolkata:20210416T171500
DTEND;TZID=Asia/Kolkata:20210416T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1127
DTSTAMP:20230914T125951Z
SUMMARY:Singly Connected Vertex Deletion Problem
DESCRIPTION:Speaker: Avinandan Das (Institut de Recherche en Informatique F
ondamentale\, Paris)\n\nAbstract: \nIn this talk\, I am going to present t
he Singly Connected Vertex Deletion Problem (SCVD).\nA digraph D is singly
connected if for all ordered pairs of vertices u\, v ∈ V (D)\, there is
at most one path in D from u to v. In this paper\, we study the Singly Co
nnected Vertex Deletion (SCVD) problem: Given an n-vertex digraph D and a
positive integer k\, does there exist a set S ⊆ V (D) such that |S| ≤
k and D − S is singly connected? This problem may be seen as a directed
counterpart of the (Undirected) Feedback Vertex Set problem\, as an undire
cted graph is singly connected if and only if it is acyclic. SCVD is known
to be NP-hard on general digraphs. We study the complexity of SCVD on var
ious classes of digraphs such as tournaments\, and various generalisations
of tournaments such as digraphs of bounded independence number\, in- and
out-tournaments and local tournaments. We show that unlike the Feedback Ve
rtex Set on Tournaments (FVST) problem\, SCVD is polynomial-time solvable
on tournaments. In addition\, we show that SCVD is polynomial-time solvabl
e on digraphs of bounded independence number\, and on the class of acyclic
local tournaments. We also study the parameterized complexity of SCVD\, w
ith k as the parameter\, on the class of in-tournaments. And we show that
on in-tournaments\, SCVD admits a fixed-parameter tractable algorithm and
a quadratic kernel.\n\nWe also show that on the class of local tournaments
\, which is a sub-class of in-tournaments\, SCVD admits a linear kernel.\n
\nZoom link: https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt
0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1127
DTSTART;TZID=Asia/Kolkata:20210423T171500
DTEND;TZID=Asia/Kolkata:20210423T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1128
DTSTAMP:20230914T125951Z
SUMMARY:A Tale of Turing Machines\, Quantum-Entangled Particles\, and Opera
tor Algebras
DESCRIPTION:Speaker: Henry Yuen (University of Toronto)\n\nAbstract: \nBelo
w event would be a screening of a past talk by Henry Yuen (available on Yo
uTube) with the same title.\n\nIn a recent result known as "MIP* = RE\," i
deas from three disparate fields of study — computational complexity the
ory\, quantum information\, and operator algebras — have come together t
o simultaneously resolve long-standing open problems in each field\, inclu
ding a 44-year old mystery in mathematics known as Connes’ Embedding Pro
blem. In this talk\, I will describe the evolution and convergence of idea
s behind MIP* = RE: it starts with three landmark discoveries from the 193
0s (Turing’s notion of a universal computing machine\, the phenomenon of
quantum entanglement\, and von Neumann’s theory of operators)\, and end
s with some of the most cutting-edge developments from theoretical compute
r science and quantum computing.\n\nThis talk is aimed at a general scient
ific audience\, and will not assume any specialized background in complexi
ty theory\, quantum physics\, or operator algebras.\n\nZoom link: https:/
/zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1128
DTSTART;TZID=Asia/Kolkata:20210430T171500
DTEND;TZID=Asia/Kolkata:20210430T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1129
DTSTAMP:20230914T125951Z
SUMMARY:Eluder dimension and sample complexity of Optimistic exploration
DESCRIPTION:Speaker: Sushant Vijayan\n\nAbstract: \nI will present the 2013
NIPS paper by Dan Russo and Van Roy where they introduce the notion of El
uder dimension and use it to analyse the UCB and Thompson Sampling algorit
hms.\n\nZoom link: https://zoom.us/j/98132227553?pwd=K2cyQllKVjExdUhlRm0vc
0ZHcEt0Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1129
DTSTART;TZID=Asia/Kolkata:20210507T171500
DTEND;TZID=Asia/Kolkata:20210507T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1130
DTSTAMP:20230914T125951Z
SUMMARY:Communication Complexity and Quantum Optimization Lower Bounds via
Query Complexity
DESCRIPTION:Speaker: Suhail Sherif\n\nAbstract: \nThe query model is a simp
le model of computation that has led to many deep results. One such class
of results relates computational complexity measures such as query complex
ity/communication complexity to algebraic measures such as degree/rank. Th
e first part of this thesis deals with such relations. The results of this
part of the thesis are given below.\n\n1. In the context of randomized co
mmunication complexity and randomized parity decision tree (RPDT) complexi
ty\, we prove that the relevant computational measures and algebraic measu
res are not closely related. This disproves multiple long-standing conject
ures in communication complexity.\n\n2. We try to quantitatively strengthe
n the above result. We do manage to do so in the context of RPDTs\, but th
ere are challenges in translating this strengthening to the context of ran
domized communication complexity. We pose a fundamental conjecture that wo
uld imply that the strengthening holds in the communication world as well.
\n\nIn the second part of the thesis\, we look at convex optimization. The
goal is to optimize a convex function in a bounded region when you can on
ly learn about the function through function value and gradient queries. W
e want to use as few queries as possible. Projected gradient descent (PGD)
is a well-known algorithm that optimally solves this task when the dimens
ion of the function's domain is large. In the second part of the thesis we
add to the optimality results of PGD.\n\n3. (a) In the case of quantum al
gorithms it was open whether there is a quadratic speedup over PGD when th
e dimension is large. We show that this is not so\, that PGD is optimal up
to constant factors.\n(b) In the case of randomized algorithms we show a
simple argument bettering the range of dimensions where PGD is known to be
optimal up to constant factors.\n\nWe end with a few open problems.\n\nJo
in Zoom Meeting\nhttps://zoom.us/j/93580606744?pwd=L1pCek1pSy91a0JOOXpXdjI
0Z2EyQT09\nMeeting ID: 935 8060 6744\nPasscode: 335095\n
URL:https://www.tcs.tifr.res.in/web/events/1130
DTSTART;TZID=Asia/Kolkata:20210517T170000
DTEND;TZID=Asia/Kolkata:20210517T180000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1131
DTSTAMP:20230914T125951Z
SUMMARY:Colloquia on the 2021 Abel Prize in Mathematics
DESCRIPTION:Speaker: Prof. Arkadev Chattopadhayay (TIFR)\, Prof. Prahladh H
arsha (TIFR)\, Prof. Mahan Maharaj (TIFR)\, Prof. Hariharan Narayanan (TIF
R)\, Prof. Jaikumar Radhakrishnan (TIFR) (School of Technology and Compute
r Science\,\nTIFR\, Mumbai.)\n\nAbstract: \nCovering Contributions of Abel
Laureate Avi Wigderson\n\nArkadev Chattopadhyay "Games Avi Plays To Prove
Hardness"\n\nPrahlad Harsha "How Avi copes with Difficulty: A computation
al perspective on randomness and knowledge"\n\nCovering Contributions of A
bel Laureate László Lovász\n\nMahan Maharaj "Graph limits"\n\nHariharan
Narayanan "Volume Computation for Convex bodies"\n\nJaikumar Radhakrishna
n "The Lovasz Local Lemma"\n\nAt Online through ZOOM webinar ( Zoom link:
https://zoom.us/j/97963259354?pwd=ZFZsa2xqWGJSZW5pUjZPNkNqeGlEZz09 )\nMeet
ing ID: 979 6325 9354 Pass code: 04072020\n
URL:https://www.tcs.tifr.res.in/web/events/1131
DTSTART;TZID=Asia/Kolkata:20210519T160000
DTEND;TZID=Asia/Kolkata:20210519T173000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1132
DTSTAMP:20230914T125951Z
SUMMARY:Hitting Sets for Orbits of Circuit Classes
DESCRIPTION:Speaker: Vishwas Bhargava (Ph.D. student at Rutgers)\n\nAbstrac
t: \nThe \\emph{orbit} of an n-variate polynomial f(\\var x) over a field
\\F\, denoted by \\orbit{f}\, is the set of polynomials obtained by applyi
ng invertible affine transformations on the variables of f(\\var x)\, and
the orbit of a polynomial class is the union of orbits of all the polynomi
als in it.\n\nIn this talk\, we will discuss recent progress on designing
hitting sets for orbits of various circuit classes. In particular\, we wil
l discuss improved construction of hitting-sets for the orbit of read-once
oblivious algebraic branching programs (ROABPs).\n\nJoint work with Sumat
a Ghosh (IIT-B).\n\nZoom link:\nhttps://zoom.us/j/93889521556?pwd=eEFJWVRt
RHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1132
DTSTART;TZID=Asia/Kolkata:20210528T171500
DTEND;TZID=Asia/Kolkata:20210528T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1133
DTSTAMP:20230914T125951Z
SUMMARY:Communication with Adversary Identification in Byzantine Multiple A
ccess channels
DESCRIPTION:Speaker: Neha Sangwan\n\nAbstract: \nIn this talk\, I will intr
oduce the problem of determining the identity of a byzantine user (interna
l adversary) in a communication system. We will consider a two-user discre
te memoryless multiple access channel where either user may deviate from t
he prescribed behaviour. Owing to the noisy nature of the channel\, it may
be overly restrictive to attempt to detect all deviations. In our formula
tion\, we only require detecting deviations which impede the decoding of t
he non-deviating user's message. When neither user deviates\, correct deco
ding is required. When one user deviates\, the decoder must either output
a pair of messages of which the message of the non-deviating user is corre
ct or identify the deviating user. The users and the receiver do not share
any randomness. The results include a characterization of the set of chan
nels where communication is feasible\, and an inner and outer bound on the
\ncapacity region.\nThe talk should be easy to follow for everyone\, inclu
ding the first years.\nZoom link:\nhttps://zoom.us/j/93889521556?pwd=eEFJW
VRtRHNpNlpZWmhNYTJGQTF6Zz09\n \n
URL:https://www.tcs.tifr.res.in/web/events/1133
DTSTART;TZID=Asia/Kolkata:20210604T171500
DTEND;TZID=Asia/Kolkata:20210604T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1134
DTSTAMP:20230914T125951Z
SUMMARY:Hitting Sets for Algebraic Models: Constructions and Consequences
DESCRIPTION:Speaker: Anamay Tengse\n\nAbstract: \nWe study hitting sets for
polynomials computed by several algebraic models. For a class of polynomi
als C\, hitting sets for C capture the problem of deterministic blackbox P
IT for C: checking if a polynomial f from C is zero by querying f on a few
points. Formally\, H is a hitting set for a class C\, if for every nonzer
o f in C\, there is some point h in H for which f(h) is nonzero. Owing to
its close connections to the search of explicit hard polynomials\, finding
small\, explicit hitting sets for various classes of polynomials is a cen
tral question in algebraic complexity theory.\n\nOur contributions towards
the study of hitting sets are as follows. We provide two explicit constru
ctions of hitting sets: quasipolynomial sized hitting sets for 'UPT circui
ts'\, and poly-sized hitting sets for log-variate 'depth-3-powering circui
ts'. Next\, we show that for any general enough algebraic model\, like cir
cuits or formulas\, even a mild improvement to the trivial hitting sets fo
r the model leads to almost efficient\, explicit hitting sets for that mod
el. Lastly\, we explore how non-trivial hitting sets for a class of polyno
mials help us in proving lower bounds against that class.\n\nThe talk will
be based on my joint works with Prerona Chatterjee (TIFR)\, Mrinal Kumar
(IITB)\, C. Ramya (TIFR) and Ramprasad Saptharishi (TIFR).\n\nZoom link: h
ttps://zoom.us/j/92326426336?pwd=V2h0WGpUeklMd01Mak1sVGpvTm9QUT09\n
URL:https://www.tcs.tifr.res.in/web/events/1134
DTSTART;TZID=Asia/Kolkata:20210609T143000
DTEND;TZID=Asia/Kolkata:20210609T153000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1135
DTSTAMP:20230914T125951Z
SUMMARY:Arithmetic Circuit Complexity of Division and Truncation
DESCRIPTION:Speaker: Gorav Jindal (TU Berlin\, Germany.)\n\nAbstract: \nGiv
en n-variate polynomials f\,g\,h such that f=g/h\, where both g and h are
computable by arithmetic circuits of size s\, we show that f can be comput
ed by a circuit of size poly(s\, deg(h)). This solves a special case of di
vision elimination for high-degree circuits (Kaltofen'87 & WACT'16). This
result is an exponential improvement over Strassen's classic result (Stras
sen'73) when deg(h) is poly(s) and deg(f) is exp(s)\, since the latter giv
es an upper bound of poly(s\, deg(f)).\nThe second part of this work deals
with the complexity of computing the truncations of uni-variate polynomia
ls or power series. We first show that the truncations of rational functio
ns are easy to compute. We also prove that the truncations of even very
simple algebraic functions are hard to compute\, unless integer factoring
is easy.\nThis is a joint work with Pranjal Dutta\, Anurag Pandey and Amit
Sinhababu. A pre-print can be found at https://eccc.weizmann.ac.il/report
/2021/072/ .\nZoom Link: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlp
ZWmhNYTJGQTF6Zz09\n \n
URL:https://www.tcs.tifr.res.in/web/events/1135
DTSTART;TZID=Asia/Kolkata:20210611T171500
DTEND;TZID=Asia/Kolkata:20210611T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1136
DTSTAMP:20230914T125952Z
SUMMARY:Approximate Polymorphisms
DESCRIPTION:Speaker: Nitin Saurabh (Technion-IIT\, Haifa\, Israel)\n\nAbstr
act: \nA Boolean function f on n variables is called a polymorphism of ano
ther Boolean function g on m variables if their operations commute. That i
s\, for all {0\,1}-matrix Z of dimension (n x m)\, f(g(row1(Z))\, g(row2(Z
))\, ...\, g(rown(Z))) = g(f(col1(Z))\, f(col2(Z))\, ...\,f(colm(Z))). The
function f is called an approximate polymorphism if this equality holds w
ith probability close to 1 when Z is sampled uniformly.\nThe problem of ch
aracterizing the structure of exact or approximate polymorphisms appears i
n several different contexts\, namely in understanding the complexity of C
SPs\, property testing\, and social choice theory.\nIn this talk\, we will
give a characterization of exact polymorphisms\, and also show that appro
ximate polymorphisms must be close to exact polymorphisms. Our results gen
eralize the classical linearity testing result of Blum et al. as well as t
he recent AND testing result of Filmus et al.\nThis is based on a joint wo
rk with Gilad Chase\, Yuval Filmus and Dor Minzer.\nZoom link: https://zoo
m.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1136
DTSTART;TZID=Asia/Kolkata:20210618T171500
DTEND;TZID=Asia/Kolkata:20210618T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1137
DTSTAMP:20230914T125952Z
SUMMARY:Machine Learning: What is it and Why should we care?
DESCRIPTION:Speaker: Preethi Jyothi (Indian Institute of Technology Bombay\
nMumbai\, Maharashtra.)\n\nAbstract: \nArtificial Intelligence\, specifica
lly machine learning (ML)\, has increasingly been making inroads into soci
ety and our lives over the last decade. We are entering an era where users
are likely to take advantage of ML-driven technologies in all walks of li
fe\, by interacting seamlessly with digital systems. But\, what is machine
learning? What are the main milestones achieved in the last decade? What
are the challenges involved with building machine learning systems? I will
aim to throw some light on these questions and also describe how ML shape
s my research on speech and language.\nYoutube link: https://youtu.be/OGKW
DcJNF7w\n
URL:https://www.tcs.tifr.res.in/web/events/1137
DTSTART;TZID=Asia/Kolkata:20210622T180000
DTEND;TZID=Asia/Kolkata:20210622T190000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1138
DTSTAMP:20230914T125952Z
SUMMARY:Great Ideas in Coding Theory: From Theory to Practice
DESCRIPTION:Speaker: Lalitha Vadlamani (International Institute of Informat
ion Technology Hyderabad\nHyderabad\, Telangana.)\n\nAbstract: \nIn 1948\,
Claude Shannon wrote his landmark paper on "A mathematical theory of comm
unication"\, which paved way to the field of information theory. He introd
uced the term entropy to measure information and also determined the funda
mental limits of source compression and communication over a noisy channel
(known as channel capacity). In his work\, he proved the existence of cod
es which achieve the channel capacity. In 1950\, Richard Hamming discovere
d the first-ever error correcting code which is single-error correcting. C
oding theorists have contributed to designing codes which achieve the chan
nel capacity in the decades to come. Low density parity check (LDPC) codes
invented by Gallager in 1960s and rediscovered in 1990s proved to be very
effective codes due to their low decoding complexity requirements at larg
e block lengths. In 2009\, Erdal Arikan proposed polar codes which are the
first class of codes provably capacity achieving for a class of binary in
put discrete memoryless channels. There are several applications of coding
theory\, apart from channel coding\, including storage devices\, distribu
ted storage systems etc. Reed-Solomon and BCH codes have been conventional
ly used in storage devices. More recently in 2012\, locally repairable cod
es (LRC) were introduced by researchers in Microsoft to solve the repair p
roblem in distributed storage systems.\nYoutube link: https://youtu.be/_Jp
MoMdX7W4\n
URL:https://www.tcs.tifr.res.in/web/events/1138
DTSTART;TZID=Asia/Kolkata:20210624T180000
DTEND;TZID=Asia/Kolkata:20210624T190000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1139
DTSTAMP:20230914T125952Z
SUMMARY:Leveraging the Invariance Principle for Out-of-Distribution General
ization
DESCRIPTION:Speaker: Karthikeyan Shanmugan (IBM Research AI\nT.J. Watson Ce
nter\, NY.)\n\nAbstract: \nOne of the fundamental issues facing deployment
of supervised learning models in real life applications is the issue of o
ut-of-distribution (OOD) generalization. Models trained using the standard
Empirical Risk Minimization (ERM) on multiple training data sources suffe
r from fitting to spurious features that correlate with label which does n
ot hold in unseen test environments. ERM’s sole focus on optimizing aver
age risk contributes to this problem. Invariance Principle\, in Pearlian C
ausal Models\, has long been used to infer causal relationships from inter
ventional data.\nInvariant Risk Minimization (IRM) is a recent paradigm th
at proposes to leverage the invariance principle in an optimization framew
ork for OOD problems. This paradigm views different training distributions
and the unseen test as intervened versions of a common but unknown causal
model. IRM seeks to identify that transformation of data such that the cl
assifier trained on top of it is invariant across training domains apart f
rom optimizing risk. Due to a challenging bilevel optimization\, a previou
s proposal was limited to handling linear classifiers. We propose a novel
game theoretic learning paradigm – called Ensemble Invariant Risk Minimi
zation (EIRM Game) whose Nash Equilibria is provably equivalent to invaria
nt solutions for a very general class of non-linear classifiers and transf
ormations. For least squares regression under unobserved confounding\, wit
h a modified game we provide the first convergence guarantees\, known for
this problem in any setting\, to approximate invariant solutions (this par
t may be discussed if time permits).\n\nBio: Karthikeyan Shanmugam is a Re
search Staff Member with the IBM Research AI group in NY in the Trustworth
y AI Department since 2017. Previously\, he was a Herman Goldstine Postdoc
toral Fellow in the Mathematical Sciences Division at IBM Research\, NY. H
e obtained his Ph.D. in Electrical and Computer Engineering from UT Austin
in 2016\, MS degree in Electrical Engineering from USC in 2012 and B.Tech
\, M.Tech degrees in Electrical Engineering from IIT Madras in 2010.\nHis
research interests broadly lie in Statistical Machine Learning (ML)\, Opti
mization\, Graph Algorithms\, and Information Theory. In ML\, his focus is
on causal inference\, online learning\, transfer learning and explainable
ML. He has won several awards in IBM for his contributions to explainable
AI and Causal Inference including the Corporate Technical Award in 2021\,
the highest technical award in IBM. His works have appeared regularly in
top AI/ML venues like NeurIPS\, ICML\, AISTATS and ICLR.\n
URL:https://www.tcs.tifr.res.in/web/events/1139
DTSTART;TZID=Asia/Kolkata:20210628T090000
DTEND;TZID=Asia/Kolkata:20210628T100000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1140
DTSTAMP:20230914T125952Z
SUMMARY:Cryptography: The Jugalbandi (duet) of Structure and Randomness
DESCRIPTION:Speaker: Shweta Agrawal (Indian Institute of Technology Madras\
nChennai\, Tamil Nadu.)\n\nAbstract: \nCryptography is a beautiful branch
of theoretical computer science that seeks to provide guarantees to the ar
t of secret keeping. The questions it poses are fundamental -- does the
universe permit asymmetry of computation? It's practical utility requires
no argument -- ad-hoc security solutions repeatedly fall prey to attack an
d crime is increasingly digital. Its scientific charm lies in its deeply p
aradoxical nature – among its early successes is the ability for two str
angers to meet\, generate a secret key and communicate privately\, all of
these from within a crowd!\nIn this talk\, we will take a closer look at t
his fascinating field\, paying special attention to the many apparent para
doxes it enables. Via examples such as zero knowledge proofs\, fully homom
orphic encryption and deniable encryption\, I hope to show you how it is t
he perennial jugalbandi\, or duet between structure and randomness that en
ables these beautiful and useful constructs.\nYoutube link: https://youtu.
be/9sQRZAdrp3c\n
URL:https://www.tcs.tifr.res.in/web/events/1140
DTSTART;TZID=Asia/Kolkata:20210629T180000
DTEND;TZID=Asia/Kolkata:20210629T190000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1141
DTSTAMP:20230914T125952Z
SUMMARY:Vertex connectivity of Eulerian orientations
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nA directed graph is said t
o be k-vertex-connected if after deleting any k-1 vertices\, therer is a d
irected path from every vertex to every other vertex along the directed ed
ges. An Eulerian orientation of a graph is an orientation such that every
vertex has equal indegree and outdegree. Given an 2k-regular graph G\, we
would like to know if every Eulerian orientation of G is k-vertex connecte
d. Horsch and Szigeti showed that this property holds for complete biparti
te graphs\, line graphs of regular complete bipartite graphs\, incidence g
raphs of projective planes\, but not for most complete graphs. In this tal
k\, we will go through their proofs. We will also show that this property
does not hold for most regular complete multipartite graphs. Link to paper
: https://doi.org/10.1016/j.dam.2020.09.022. Zoom link is: https://zoom.us
/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1141
DTSTART;TZID=Asia/Kolkata:20210709T150000
DTEND;TZID=Asia/Kolkata:20210709T160000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1142
DTSTAMP:20230914T125952Z
SUMMARY:Fooling Boolean functions with expander random walks
DESCRIPTION:Speaker: Ashutosh Shankar\n\nAbstract: \nExpander graphs are sp
arse but highly connected graphs\, which find a variety of uses in CS. If
the vertices of an expander are labelled by 0 or 1\, a $t$-step walk gives
a $t$-bit string. Cohen\, Peri and Ta-Shma (2020) consider the question:
can a Boolean function on $t$ variables distinguish between a truly random
$t$-bit string and a $t$-step walk in a labelled expander? We will see th
at Majority and other symmetric functions can indeed be fooled by a walk.
However\, the "fooling" does not seem to improve with $t$\; we will see a
counterexample from a subsequent paper if time allows.\n\nZoom link : http
s://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1142
DTSTART;TZID=Asia/Kolkata:20210716T171500
DTEND;TZID=Asia/Kolkata:20210716T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1143
DTSTAMP:20230914T125952Z
SUMMARY:The Strahler Number of a Parity Game
DESCRIPTION:Speaker: Thejaswini Raghavan (University of Warwick)\n\nAbstrac
t: \nThe Strahler number of a rooted tree is the largest height of a perfe
ct binary tree that is its minor. The Strahler number of a parity game is
proposed to be defined as the smallest Strahler number of the tree of any
of its attractor decompositions. It is proved that parity games can be sol
ved in quasi-linear space and in time that is polynomial in the number of
vertices n and linear in (d/2k)^k\, where d is the number of priorities an
d k is the Strahler number. This complexity is quasi-polynomial because th
e Strahler number is at most logarithmic in the number of vertices. The pr
oof is based on a new construction of small Strahler-universal trees.\n\nI
t is shown that the Strahler number of a parity game is a robust\, and hen
ce arguably natural\, parameter: it coincides with its alternative version
based on trees of progress measures and—remarkably—with the register
number defined by Lehtinen (2018). It follows that parity games can be sol
ved in quasi-linear space and in time that is polynomial in the number of
vertices and linear in (d/2k)^k\, where k is the register number. This sig
nificantly improves the running times and space achieved for parity games
of bounded register number by Lehtinen (2018) and by Parys(2020). The runn
ing time of the algorithm based on small Strahler-universal trees yields a
novel trade-off k.log(d/k) = O(log n) between the two natural parameters
that measure the structural complexity of a parity game\, which allows sol
ving parity games in polynomial time. This includes as special cases the a
symptotic settings of those parameters covered by the results of Calude\,
Jain\, Khoussainov\, Li\, and Stephan (2017)\, of Jurdzinski and Lazic (20
17)\, and of Lehtinen (2018)\, and it significantly extends the range of s
uch settings\, for example to d = 2^{O(\\sqrt{\\lg n})} and k = O(\\sqrt{\
\lg n}).\n\nThis is joint work with Laure Daviaud and Marcin Jurdzinski.\n
\nZoom link: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF
6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1143
DTSTART;TZID=Asia/Kolkata:20210723T171500
DTEND;TZID=Asia/Kolkata:20210723T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1144
DTSTAMP:20231128T111754Z
SUMMARY:A Computational Approach towards Incentives in Social Choice
DESCRIPTION:Speaker: Rohit Vaish\n\nAbstract: \nGroup decision-making is a
ubiquitous phenomenon with diverse applications ranging from political ele
ctions to recommender systems and from organ exchanges to online marketpla
ces. Social choice is a subfield of economics that provides a formal frame
work for studying group decision-making procedures. Classically\, social c
hoice theory has focused on establishing abstract results concerning the e
xistence of procedures that provide the desired incentives to the particip
ating agents. However\, in order to be practically applicable\, the mere e
xistence of such procedures is not enough---efficient computation is impor
tant as well. In this talk\, I will illustrate the role of computation in
shaping agents' incentives via a case study in fair division. Specifically
\, I will talk about fair division of indivisible goods\, which is a relev
ant model for assigning seats in university courses\, allocating public ho
using units\, and inheritance division. I will present an algorithmic fram
ework that combines the local search paradigm with the classical Fisher ma
rket model from economics\, and simultaneously achieves the seemingly inco
mpatible goals of fairness and economic efficiency. I will conclude with a
n overview of my other work and future research directions. Bio: Rohit Vai
sh is a visiting fellow at Tata Institute of Fundamental Research (TIFR).
Previously\, he was a postdoctoral researcher at Rensselaer Polytechnic In
stitute (RPI) and\, prior to that\, received his PhD from Indian Institute
of Science (IISc). His research is in computational social choice---a rap
idly growing area at the intersection of theoretical computer science\, ar
tificial intelligence\, and economics. He has worked on problems in voting
\, matching\, fair division\, and learning theory\, and his research has b
een published in top journals like Artificial Intelligence (AIJ) and premi
er theory and AI conferences such as EC\, SODA\, AAAI\, IJCAI\, and NeurIP
S among others. In addition\, he is a recipient of Prof. R Narasimhan post
doctoral award at TIFR\, a best paper award nomination at AAMAS 2018\, and
the INSPIRE faculty fellowship.\n
URL:https://www.tcs.tifr.res.in/web/events/1144
DTSTART;TZID=Asia/Kolkata:20210726T103000
DTEND;TZID=Asia/Kolkata:20210726T113000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1145
DTSTAMP:20230914T125952Z
SUMMARY:Superpolynomial lower bounds against low-depth algebraic circuits
DESCRIPTION:Speaker: Sébastien Tavenas (Univ. Grenoble Alpes\,\nUniv. Savo
ie Mont Blanc\,\nCNRS\, LAMA.)\n\nAbstract: \nAn algebraic circuit compute
s a polynomial using addition and multiplication operators. Understanding
the power of algebraic circuits has close connections to understanding gen
eral computation. It is known that proving lower bounds for algebraic circ
uits can serve as a stepping stone towards proving general Boolean circuit
lower bounds.\nDespite this\, not many lower bounds are known for even si
mple Sigma Pi Sigma (product-depth 1) circuits. Before our work\, the best
known lower bound for product-depth 1 circuit was (slightly less than) cu
bic. No lower bounds were known for general product-depth 2 circuits.\nIn
this work\, we show the first superpolynomial lower bound for low-product-
depth algebraic circuits.\nIn the talk\, we discuss the main results and p
resent the proof ideas used in the proof of the superpolynomial lower boun
d for product-depth 1 circuits.\n\nThis talk is based on joint work with N
utan Limaye and Srikanth Srinivasan.\nYouTube link : https://youtu.be/S1_k
4OxC1bE\n
URL:https://www.tcs.tifr.res.in/web/events/1145
DTSTART;TZID=Asia/Kolkata:20210727T160000
DTEND;TZID=Asia/Kolkata:20210727T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1146
DTSTAMP:20230914T125952Z
SUMMARY:Efficient Stochastic Optimisation using Black Box Importance Sampli
ng
DESCRIPTION:Speaker: Anand Deo (Singapore University of Technology and Desi
gn)\n\nAbstract: \nMitigating the effect of tail risk has gained prominenc
e in a variety of applications where safety is of paramount importance. Ta
il risk averseness is typically incorporated into standard optimisation mo
dels by penalising decisions which lead to poor tail performance\, often c
aptured through the use of measures such as Value at Risk (V@R) and Condit
ional Value at Risk (CV@R). Operationalising these risk sensitive optimisa
tion problems however\, requires accurate computation of tail expectations
of random variables and may incur a large sample requirement. In this tal
k\, we discuss the use of Black Box Importance Sampling (BBIS) to mitigate
this difficulty. Specifically\, we show that given black box access to th
e loss causing covariates\, BBIS significantly reduces the sample complexi
ty of solving a wide range of tail risk averse optimisation problems. This
differs from most of the state of the art\, where algorithms to reduce sa
mple requirements are carefully tuned to the problem at hand. The distribu
tion/loss agnostic nature of BBIS leads to a wide applicability\, ranging
from relatively simple instances such as linear portfolio optimisation to
complicated ones such as two stage problems. Numerical simulations support
our theoretical claims and help establish the utility of BBIS in a number
of practically relevant settings.\n\nThis talk is based on a number of jo
int works with Karthyek Murthy\, SUTD. I will not assume any prerequisites
beyond a basic understanding of probability. Technicalities will be kept
to the minimum\, and the exposition will be mostly pictorial.\n\nZoom link
: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1146
DTSTART;TZID=Asia/Kolkata:20210730T171500
DTEND;TZID=Asia/Kolkata:20210730T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1147
DTSTAMP:20230914T125952Z
SUMMARY:Exact Sampling & List-Decoding
DESCRIPTION:Speaker: Siddharth Bhandari\n\nAbstract: \nWe will discuss the
following results. 1. Exact sampling: We will present an efficient algorit
hm that\, given a graph of maximum degree $\\Delta$ and a list of at least
$3\\Delta+1$ colours\, produces a random colouring of the graph that is \
\emph{exactly} uniformly distributed on the set of all proper colourings.
Our algorithm can be generalized to the setting of list colourings\, where
each vertex is provided with a separate list of at least $3\\Delta+1$ col
ours. Before this work\, it was known that exact sampling was possible if
about $\\Delta^2$ colours were allowed. 2. List-decoding error-correcting
codes: In list-decoding\, the decoder\, based on the received word\, is re
quired to output a small list of messages\, one of which must be the origi
nal message. We will discuss the following results about list-decoding. (a
) Zero-error list decoding capacity of the $q/(q-1)$ channel: We will pres
ent a lower bound showing that the zero-error list decoding capacity of th
is channel is exponentially small in $q$ even if the list size is allowed
grow as $\\frac{1}{6} q \\ln q$. Previous results showed that the capacity
was exponentially small if the list size was allowed to grow no larger th
an $1.58q$. (b) Multiplicity codes: We consider a natural generalization o
f Reed-Muller codes where at each evaluation point\, one records not only
the evaluation of the message polynomial\, but also all its partial deriva
tives up to a certain order. We will present an efficient algorithm that (
under mild assumptions) list-decodes multivariate multiplicity codes on ar
bitrary grids up to their distance. Previously such results were known onl
y for univariate multiplicity codes. (c) Polynomial ideal codes: A polynom
ial ideal code is specified by a collection of relatively prime monic poly
nomials. The encoding is obtained by specifying the remainders of the mess
age polynomial modulo the various polynomials in the collection. We will d
escribe an efficient algorithm to list-decode some special polynomial idea
l codes\, which we \\emph{call affine folded Reed-Solomon codes\,} up to t
heir distance. Previous results allowed list-decoding up to the distance f
or folded Reed-Solomon codes\, univariate multiplicity codes and additive
folded Reed-Solomon codes\, which are all instances of affine folded Reed-
Solomon codes. (The above results were obtained in collaborations that inv
olved the following: Sayantan Chakraborty\, Prahladh Harsha\, Mrinal Kumar
\, Jaikumar Radhakrishnan\, Madhu Sudan.)\n
URL:https://www.tcs.tifr.res.in/web/events/1147
DTSTART;TZID=Asia/Kolkata:20210803T160000
DTEND;TZID=Asia/Kolkata:20210803T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1148
DTSTAMP:20230914T125952Z
SUMMARY:Regret minimization in heavy-tailed bandits.
DESCRIPTION:Speaker: Shubhada Agrawal\n\nAbstract: \nIn this talk\, we will
revisit the classical regret-minimization problem in the multi-armed band
it setting. This problem has been well studied when the arm distributions
are restricted to have either bounded support or belong to a single parame
ter exponential family (distributions characterized by 1 parameter\, for e
xample\, Gaussian with fixed variance\, Bernoulli distributions\, Poisson
distributions\, etc.). However\, in many applications\, the underlying ar
m distributions fail to satisfy these simplifying assumptions. We will con
sider a much general class of arm distributions and work with a weaker ass
umption that the moments of order $(1 + \\epsilon)$ are uniformly bounded
by a known constant B\, for some given $\\epsilon > 0$. We will look at an
optimal algorithm that matches the lower bound exactly in the in the firs
t-order term.\n\nRegret-minimization algorithms typically rely on construc
ting tight confidence intervals for means of the underlying distributions.
We will also look at new anytime-valid confidence intervals for means\, w
hich are based on the empirical likelihood principle. Algorithms using the
MGF-based concentration of the empirical mean for bounded support distrib
utions (Auer et al.\, 2002)\, or of the robust estimators for mean\, like
the truncated empirical mean in the case of heavy-tailed distributions (Bu
beck et al.\, 2013)\, exist in the literature. We will see exactly where t
he framework of the optimal algorithm gains over these existing algorithms
that use MGF-based confidence intervals for the mean.\n\nIf time permits\
, we will also look at a mixture-martingale-based proof for the validity o
f the proposed confidence intervals.\n\nThis talk is based on joint work w
ith Sandeep Juneja and Wouter M. Koolen (the paper can be found here). I w
ill not assume any prerequisites beyond a basic understanding of probabili
ty theory.\n\nZoom link: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlp
ZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1148
DTSTART;TZID=Asia/Kolkata:20210806T171500
DTEND;TZID=Asia/Kolkata:20210806T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1149
DTSTAMP:20230914T125952Z
SUMMARY:Fourier Analytic Techniques in Theoretical Computer Science
DESCRIPTION:Speaker: Somnath Chakraborty\n\nAbstract: \nIn this talk\, we w
ill discuss the following problems.\n\nProblem 1\n"Suppose that $M$ is unk
nown finite point set in $\\mathbb R^d$. Suppose $\\omega$ is some distrib
ution on $M$\, and $\\gamma$ is the standard $d$-dimensional Gaussian. Let
$X_\\omega$ and $X_\\gamma$ be independent random variables whose distrib
utions are $\\omega$ and $\\gamma$\, respectively. Can we devise an effici
ent randomized algorithm that\, given small $\\epsilon\,\\delta>0$\, and i
ndependent random samples of $X_\\omega+X_\\gamma$\, outputs a distributio
n $\\hat\\omega$ such that support of $\\hat\\omega$ is a finite point set
$\\hat M$\, having same cardinality as $M$\, and within distance $\\epsil
on$ from $M$\, and $\\sup_{m\\in M}|\\omega(m)-\\hat\\omega(\\pi(m))|$ is
`small' for some bijection $\\pi:M\\rightarrow\\hat M$\, with success prob
ability at least $1-\\delta$?"\nProblem 2\n"Suppose that $M$ is a sphere o
f dimension $d$\, and $\\mbox{Lip}_1(M)$ consists of all the Lipschitz fun
ctions defined on $M$\, having Lipschitz constant at most 1. Can we devise
an efficient randomized algorithm that\, given small $\\epsilon\,\\delta>
0$\, for any input $\\phi\\in \\mbox{Lip}_1(M)$\, outputs a function $\\ha
t\\phi$ on $M$\, such that $||\\hat\\phi-\\phi||_1\n
URL:https://www.tcs.tifr.res.in/web/events/1149
DTSTART;TZID=Asia/Kolkata:20210811T110000
DTEND;TZID=Asia/Kolkata:20210811T120000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1150
DTSTAMP:20230914T125952Z
SUMMARY:Online Energy Minimization Under A Peak Age of Information Constrai
nt
DESCRIPTION:Speaker: Kumar Saurav\n\nAbstract: \nWe consider a node where p
ackets of fixed size are generated at arbitrary intervals. The node is req
uired to maintain the peak age of information (AoI) at the monitor below a
threshold by transmitting potentially a subset of the generated packets.
At any time\, depending on packet availability and current AoI\, the node
can choose the packet to transmit\, and its transmission speed. We conside
r a power function (rate of energy consumption) that is increasing and con
vex in transmission speed\, and the objective is to minimize the energy co
nsumption under the peak AoI constraint at all times. For this problem\, w
e propose a (customized) greedy policy\, and analyze its competitive ratio
(CR) by comparing it against an optimal offline policy by deriving some s
tructural results. We show that for polynomial power functions\, the CR up
per bound for the greedy policy is independent of the system parameters\,
such as the peak AoI\, packet size\, time horizon\, or the number of packe
ts generated. Also\, we derive a lower bound on the competitive ratio of a
ny causal policy\, and show that for exponential power functions (e.g.\, S
hannon rate function)\, the competitive ratio of any causal policy grows e
xponentially with increase in the ratio of packet size to peak AoI.\n\nThi
s talk is based on the joint work with Prof. Rahul Vaze\, set to appear in
the Proc. WiOpt 2021.\n\nZoom link: https://zoom.us/j/93889521556?pwd=eEF
JWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1150
DTSTART;TZID=Asia/Kolkata:20210813T171500
DTEND;TZID=Asia/Kolkata:20210813T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1151
DTSTAMP:20230914T125952Z
SUMMARY:Criticality of boolean functions and Entropy Switching lemma for DN
Fs
DESCRIPTION:Speaker: Tulasi mohan Molli\n\nAbstract: \nA p-random restricti
on is a random partial input obtained by independently leaving each input
variable unset with probability p setting them to either 0\,1 with (1-p)/2
each.\n\nCriticality of a boolean function f is the inverse of the probab
ility p at which the decision tree depth t of a random restriction of f go
es down exponentially in t.The ground breaking work of Hastad’s Switchin
g lemma is a bound on criticality of CNFs and DNFs. Rossman defined the no
tion of criticality and showed its connection to average case lower bounds
\, fourier tail bounds\, and decision tree size.\n\nIn this talk we will s
ee implications of bounds on criticality to average case lower bounds\, Fo
urier tail bounds and decision tree size and an Entropy Switching lemma fo
r DNFs due to Rossman.\n\nZoom link:\nhttps://zoom.us/j/93889521556?pwd=eE
FJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1151
DTSTART;TZID=Asia/Kolkata:20210820T171500
DTEND;TZID=Asia/Kolkata:20210820T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1152
DTSTAMP:20230914T125952Z
SUMMARY:Collapses and Persistent Homology
DESCRIPTION:Speaker: Siddharth Pritam (DataShape\, Inria)\n\nAbstract: \nWe
introduce two new approaches to compute the Persistent Homology (PH) of a
sequence of simplicial complexes. The basic idea is to simplify the compl
exes of the input sequence by using special types of collapses (strong and
edge collapse) and to compute the PH of an induced sequence of smaller si
ze that has the same PH as the initial one. Our first approach uses strong
collapse which is introduced by J. Barmak and E.Miniam [DCG (2012)]. Stro
ng collapse consists of removal of special vertices called dominated verti
ces from a simplicial complex. In the second approach\, we extend the noti
ons of dominated vertex to a simplex of any dimension. Domination of edges
appears to be very powerful and we study it in the case of flag complexes
in more detail. As a result and as demonstrated by numerous experiments o
n publicly available data sets\, our approaches are extremely fast and mem
ory efficient in practice.\n
URL:https://www.tcs.tifr.res.in/web/events/1152
DTSTART;TZID=Asia/Kolkata:20210824T160000
DTEND;TZID=Asia/Kolkata:20210824T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1153
DTSTAMP:20230914T125952Z
SUMMARY:An Asymptotic Analysis of Risk in Financial Systems - Analysis and
Algorithms
DESCRIPTION:Speaker: Anand Deo\n\nAbstract: \nOver the past few decades\, p
robabilistic models have become an important tool for under-standing risks
and decision making in practical ﬁnancial systems. In the design of suc
h systems one often wishes to relate the risk to the statistics of underly
ing stochasticity. However\, this task is complicated by the fact that rea
listic ﬁnancial systems are complex\, and undesirable events in them are
often rare. A large body of research has been devoted to understanding th
e nature of such rare events\, and how they relate to the stochasticity a
ﬀecting the system. In this talk\, we undertake a detailed study of thes
e aspects in order to develop structural insights on a number of ﬁnancia
l systems of practical interest. Our main contributions are as below:\nI.
We discuss the development of a closed form\, interpretable parameter esti
mation technique for predicting defaults of ﬁnancial ﬁrms. Typically\,
one uses maximum likelihood estimation (MLE) for predicting the ﬁrm def
ault probabilities. We prove that our estimator is almost as accurate as t
he MLE\, verify our result empirically on a sample of US corporate data\,
and showcase the computational/interpretative beneﬁts of our estimator o
ver the MLE.\nII. We develop a statistically consistent estimator for cond
itional value-at-risk (CVaR) based optimization objectives and their gradi
ents. Unlike the state-of-the-art sample average approximations\, the prop
osed approximation scheme exploits the self-similarity of heavy-tailed dis
tributions to extrapolate data from lower quantiles\, thereby reducing dat
a requirements for accurate estimation.\nIII. Motivated by the increasing
adoption of models which facilitate automation in risk management and deci
sion-making\, we present a novel importance sampling (IS) scheme for measu
ring distribution tails of objectives. Conventional eﬃcient IS approache
s suﬀer from feasibility concerns due to the need to intricately tailor
the sampler to the underlying probability distribution and the objective.
We overcome this challenge in the proposed black-box scheme by automating
the selection of an eﬀective IS distribution with a transformation that
implicitly learns and replicates the concentration properties observed in
less rare samples.\nIV. We develop a limiting representation for an interc
onnected banking network in presence of partial information. Practical ban
king networks are large and complicated\, and one searches for simple limi
ting representations (as the network size goes to inﬁnity). We character
ise the wealth of banks in a large network in terms of a simple\, one dime
nsional distributional ﬁxed point\, which we show is amenable to Monte C
arlo simulation.\nThis talk is based on joint work with Sandeep Juneja and
Karthyek Murthy.\n\nThe zoom link for the talk is https://zoom.us/j/93128
173558?pwd=SXRkdFE1MVBnc2hSSEtvbHRIZG4yQT09\n
URL:https://www.tcs.tifr.res.in/web/events/1153
DTSTART;TZID=Asia/Kolkata:20210825T183000
DTEND;TZID=Asia/Kolkata:20210825T193000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1154
DTSTAMP:20230914T125952Z
SUMMARY:Paper Presentation
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nPaper: Nachum Dershowitz a
nd Yuri Gurevich\, "A Natural Axiomatization of Computability and Proof of
Church's Thesis"\nhttps://doi.org/10.2178/bsl/1231081370\n
URL:https://www.tcs.tifr.res.in/web/events/1154
DTSTART;TZID=Asia/Kolkata:20210827T100000
DTEND;TZID=Asia/Kolkata:20210827T110000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1155
DTSTAMP:20230914T125952Z
SUMMARY:Paper Presentation
DESCRIPTION:Speaker: Hari Krishnan P A\n\nAbstract: \n"Paper: Yuchen Zhang\
, John C Duchi\, Michael I Jordan\, Martin J Wainwright\, Information-theo
retic lower bounds for distributed statistical estimation with communicati
on constraints"\nhttps://proceedings.neurips.cc/paper/2013/hash/d6ef5f7fa9
14c19931a55bb262ec879c-Abstract.html\n
URL:https://www.tcs.tifr.res.in/web/events/1155
DTSTART;TZID=Asia/Kolkata:20210827T110000
DTEND;TZID=Asia/Kolkata:20210827T120000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1156
DTSTAMP:20230914T125952Z
SUMMARY:Paper Presentation
DESCRIPTION:Speaker: Praveen Sai Chinthaginjala\n\nAbstract: \nPaper: Nikhi
l R. Devanur\, Zhiyi Huang. "Primal Dual Gives Almost Optimal Energy-Effic
ient Online Algorithms"\nhttps://dl.acm.org/doi/abs/10.1145/3155297\n
URL:https://www.tcs.tifr.res.in/web/events/1156
DTSTART;TZID=Asia/Kolkata:20210827T120000
DTEND;TZID=Asia/Kolkata:20210827T130000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1157
DTSTAMP:20230914T125952Z
SUMMARY:Paper Presentation
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nPaper: Michael Saks an
d Rahul Santhanam\, "Circuit Lower Bounds from NP-Hardness of MCSP Under T
uring Reductions"\nhttps://drops.dagstuhl.de/opus/volltexte/2020/12578/\n
URL:https://www.tcs.tifr.res.in/web/events/1157
DTSTART;TZID=Asia/Kolkata:20210827T140000
DTEND;TZID=Asia/Kolkata:20210827T150000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1158
DTSTAMP:20230914T125952Z
SUMMARY:Project: An algorithm for the multiplicity Schwartz-Zippel lemma
DESCRIPTION:Speaker: Ashutosh Shankar\n\nAbstract: \nThe multiplicity Schwa
rtz-Zippel lemma provides a bound on the total multiplicity of zeroes of a
low-degree polynomial within a product set. It implies that multiplicity
codes\, which can be thought of as generalizations of Reed-Muller codes th
at consist of evaluations of derivatives along with the evaluations of the
polynomial - have good distance. There has been progress towards the prob
lem of algorithmizing this lemma - that is\, unique decoding to find the c
losest codeword given a possibly corrupted received word - for certain cas
es such as the product set being a vector space\, or the number of derivat
ives given being large. In this project\, we describe an algorithm for the
multiplicity Schwartz-Zippel lemma for general product sets and any numbe
r of derivatives. For the purposes of this project\, we restrict ourselves
to the bivariate case. Our algorithm is motivated by Kim and Kopparty’s
algorithm for decoding Reed-Muller codes over product sets. In the proces
s of adapting it\, we develop new notions of weight and distance\, as well
as a new approach to analysing Forney’s classical algorithm for decodin
g concatenation codes. The results presented in this work are based on joi
nt work with Siddharth Bhandari\, Prahladh Harsha and Mrinal Kumar (IIT Bo
mbay).\n
URL:https://www.tcs.tifr.res.in/web/events/1158
DTSTART;TZID=Asia/Kolkata:20210827T150000
DTEND;TZID=Asia/Kolkata:20210827T160000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1159
DTSTAMP:20230914T125952Z
SUMMARY:The complexity of approximating Satisfiable CSPs
DESCRIPTION:Speaker: Amey Bhangale (University of California)\n\nAbstract:
\nRaghavendra's famous result from 2008 fully characterizes the inapproxim
ability of every maximum constraint satisfaction problem (Max-CSP). Howeve
r\, the result inherently loses perfect completeness. This means that it d
oes not say anything about the complexity of solving satisfiable Max-CSP i
nstances.\n\nIn this talk\, I will discuss inapproximability of linear equ
ations over non-abelian groups. We show that even if the instance is satis
fiable\, it is NP-hard to beat the random assignment algorithm for 'perfec
t groups'. This is in stark contrast to the problem of solving linear equa
tions over abelian groups\, in which case we know how to find a satisfying
assignment\, if it exists\, efficiently. We also give tight inapproximabi
lity results for solving linear equations over any non-abelian group. \n\n
The proof techniques involve Fourier analysis over non-abelian groups and
lots of applications of the Cauchy-Schwarz inequality.\n\nThis is joint wo
rk with Subhash Khot.\n
URL:https://www.tcs.tifr.res.in/web/events/1159
DTSTART;TZID=Asia/Kolkata:20210831T160000
DTEND;TZID=Asia/Kolkata:20210831T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1160
DTSTAMP:20230914T125952Z
SUMMARY:Linear Matroid Intersection is in Pseudo-deterministic NC
DESCRIPTION:Speaker: Sumanta Ghosh (IIT Bombay --> Caltech)\n\nAbstract: \n
A pseudo-deterministic NC algorithm for a search problem is an RNC algorit
hm that\, for a given input\, outputs a fixed solution with high probabili
ty. In this talk\, we describe a pseudo-deterministic NC algorithm for the
linear matroid intersection problem. Our work strengthens the RNC algorit
hm for the linear matroid intersection given by Narayanan\, Saran and Vazi
rani (NSV'94). It also generalizes the pseudo-deterministic NC algorithm f
or the bipartite matching (due to Goldwasser and Grossman 2017) to linear
matroid intersection.\nIt is a joint work with Rohit Gurjar.\n\nZoom link:
https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1160
DTSTART;TZID=Asia/Kolkata:20210903T171500
DTEND;TZID=Asia/Kolkata:20210903T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1161
DTSTAMP:20230914T125952Z
SUMMARY:Quasi-transitive tilings of the plane
DESCRIPTION:Speaker: Arun Maiti (Indian Institute of Science\, Bangalore.)\
n\nAbstract: \nTiling of a surface is acted upon by its automorphism group
. The tilings with single vertex orbit\, the transitive tilings\, are wel
l studied since antiquity. In this talk I will present a class of locally
symmetric tilings of the plane that has finitely many orbits but not trans
itive. Along the way I will talk about some known and a few new results ab
out locally symmetric and pseudo-symmetric tilings of the plane.\n
URL:https://www.tcs.tifr.res.in/web/events/1161
DTSTART;TZID=Asia/Kolkata:20210907T160000
DTEND;TZID=Asia/Kolkata:20210907T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1162
DTSTAMP:20230914T125953Z
SUMMARY:Commuting Matrices and Multivariate Multiplicity
DESCRIPTION:Speaker: Anamay Tengse\n\nAbstract: \nWe know that for matrices
A and B\, AB is not the same as BA in general. But suppose B is a polynom
ial in A\, like B = A^2 - 3A + I (note I = A^0). Then AB is indeed the sam
e as BA. Some natural questions follow for a set S of (square) matrices th
at commute with each other.\n1. Does there always exist an A in S that "ge
nerates" the rest via polynomials?\n2. Since diagonal matrices commute wit
h each other\, is it the case that we can (simultaneously) diagonalise all
matrices in S?\n3. If (1) and (2) are false\, then what do these non-triv
ial sets S look like?\nIn this talk we will see an elegant characterisatio
n of commuting matrices which follows from the works of Marinari\, Möller
and Mora (1993)\, and Möller and Stetter (1995). Similar in spirit to qu
estion (1) above\, the characterisation involves multivariate polynomials
and uses a notion of "multiplicity of a polynomial at a point" that is sli
ghtly non-standard in CS.\n\nZoom link:\nhttps://zoom.us/j/93889521556?pwd
=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n \n
URL:https://www.tcs.tifr.res.in/web/events/1162
DTSTART;TZID=Asia/Kolkata:20210917T171500
DTEND;TZID=Asia/Kolkata:20210917T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1163
DTSTAMP:20230914T125953Z
SUMMARY:Resource Allocation Problems in Large-Scale Networks
DESCRIPTION:Speaker: Abhishek Sinha (Indian Institute of Technology Madras)
\n\nAbstract: \nOptimal resource allocation in networks gives rise to some
of the most fundamental problems at the intersection of algorithms\, stoc
hastic processes\, and learning. In this talk\, we will discuss our recent
contributions to three canonical resource allocation problems\, namely ca
ching\, routing\, and scheduling. First\, we will consider the optimal cac
hing problem for single and networked caches. However\, instead of minimiz
ing the competitive ratio - the classical metric of choice for caching pro
blems\, we will look at the problem from an online learning perspective th
at minimizes regret. We will show that this viewpoint leads to an entirely
new class of caching policies with provably better performance than the c
lassical ones. We will also discuss some converse results on the regret lo
wer bounds for this problem. Next\, we will discuss the problem of through
put-optimal dynamic routing of a broad class of traffic\, including unicas
t\, multicast\, broadcast\, and anycast flows on a network with arbitrary
link scheduling constraints. We will present a unified algorithmic framewo
rk based on precedence relaxations\, leading to an efficient policy that p
rovably outperforms the state-of-the-art Backpressure routing algorithm. F
inally\, we will discuss a user scheduling problem for reliable and fresh
information delivery over unreliable wireless channels. However\, contrary
to the existing literature\, which predominantly considers stochastic cha
nnels\, we investigate a non-stationary environment modeled using a new ad
versarial framework. We will describe competitive algorithms in this setti
ng along with some approximately tight lower bounds. We will supplement ea
ch part of the talk with a set of open problems.\n\nBio: Abhishek Sinha is
currently working as an Assistant Professor in the Department of Electric
al Engineering at the Indian Institute of Technology Madras. He received h
is Ph.D. degree from the Massachusetts Institute of Technology in 2017\, w
here he was associated with the Laboratory for Information and Decision Sy
stems (LIDS). After his Ph.D.\, Abhishek worked as a senior engineer at Qu
alcomm Research\, San Diego\, in the 5G standardization group. He obtained
his M.E. degree in Telecommunication Engg. from the Indian Institute of S
cience\, Bangalore\, and his B.E. degree in Electronics and Telecommunicat
ion Engg. from Jadavpur University\, Kolkata\, India. He is a recipient of
the Best Paper Award in INFOCOM 2018\, the Best Paper Award in MobiHoc 20
16\, Prof. Jnansaran Chatterjee memorial gold medal\, and T.P. Saha Memori
al gold-centered silver medal from Jadavpur University and Jagadis Bose Na
tional Science Talent Search (JBNSTS) scholarship\, Kolkata\, India. His a
reas of interest include networks\, information theory\, theoretical machi
ne learning\, and applied probability.\n
URL:https://www.tcs.tifr.res.in/web/events/1163
DTSTART;TZID=Asia/Kolkata:20210921T160000
DTEND;TZID=Asia/Kolkata:20210921T171500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1164
DTSTAMP:20230914T125953Z
SUMMARY:Parallel Repetition for the GHZ Game: A Simpler Proof
DESCRIPTION:Speaker: Uma Girish (Princeton University)\n\nAbstract: \nWe gi
ve a new proof of the fact that the parallel repetition of the (3-player)
GHZ game reduces the value of the game to zero polynomially quickly. That
is\, we show that the value of the n-fold parallel repetition of the GHZ g
ame is at most n^{-\\Omega(1)}. This was first established by Holmgren and
Raz [HR20]. We present a new proof of this theorem that we believe to be
simpler and more direct. Unlike most previous works on parallel repetition
\, our proof makes no use of information theory\, and relies on the use of
Fourier analysis. The GHZ game [GHZ89] has played a foundational role in
the understanding of quantum information theory\, due in part to the fact
that quantum strategies can win the GHZ game with probability 1. It is pos
sible that improved parallel repetition bounds may find applications in th
is setting. Recently\, Dinur\, Harsha\, Venkat\, and Yuen [DHVY17] highlig
hted the GHZ game as a simple three-player game\, which is in some sense m
aximally far from the class of multi-player games whose behavior under par
allel repetition is well understood. Dinur et al. conjectured that paralle
l repetition decreases the value of the GHZ game exponentially quickly\, a
nd speculated that progress on proving this would shed light on parallel r
epetition for general multi-player (multi-prover) games. This is based on
a joint work with Justin Holmgren\, Kunal Mittal\, Ran Raz and Wei Zhan.\n
YouTube link : https://www.youtube.com/watch?v=jxQ81fDMpOU\n
URL:https://www.tcs.tifr.res.in/web/events/1164
DTSTART;TZID=Asia/Kolkata:20210928T190000
DTEND;TZID=Asia/Kolkata:20210928T200000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1165
DTSTAMP:20230914T125953Z
SUMMARY:The Space Complexity of Sum Labelling
DESCRIPTION:Speaker: Kshitij Gajjar (National University of Singapore)\n\nA
bstract: \nHow does one store a graph in the database? Typically the verti
ces are labelled by a set {1\, 2\, ...\, n}. The edges can be denoted in m
any different ways: adjacency matrix\, incidence matrix\, adjacency list\,
to name a few. But what if the vertices are labelled in a more creative w
ay\, such that the labels of the vertices themselves denote their adjacenc
ies? This eliminates the need for storing the edges! This topic is part of
a heavily researched field called graph labelling\, with connections to c
oding theory and information theory. In this talk\, we will explore a type
of graph labelling known as sum labelling. This is joint work with Hennin
g Fernau (https://eccc.weizmann.ac.il/report/2021/114).\n\nZoom Link: http
s://zoom.us/j/93889521556pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1165
DTSTART;TZID=Asia/Kolkata:20211001T171500
DTEND;TZID=Asia/Kolkata:20211001T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1166
DTSTAMP:20230914T125953Z
SUMMARY:Computationally Secure Computation from One-Way Noisy Communication
DESCRIPTION:Speaker: Varun Narayanan (Technion)\n\nAbstract: \nCan a sender
encode a pair of messages (m0\, m1) jointly\, and send their encoding ove
r (say) a binary erasure channel\, so that the receiver can decode exactly
one of the two messages and the sender does not know which one?\nGarg et
al. (Crypto 2015) showed that this is information-theoretically impossible
. We show how to circumvent this impossibility by assuming that the receiv
er is computationally bounded\, settling for an inverse- polynomial securi
ty error (which is provably necessary)\, and relying on ideal obfuscation.
Our solution creates a "computational anti-correlation" between the event
s of receiving m0 and receiving m1 by exploiting the anti-concentration of
the binomial distribution.\nThe ideal obfuscation primitive in our constr
uction can either be directly realized using (stateless) tamper-proof har
dware\, yielding an unconditional result.\nAs a corollary\, we get similar
feasibility results for general secure computation of sender-receiver fun
ctionalities by leveraging the completeness of the above random oblivious
transfer functionality.\n\nZoom link: \nhttps://zoom.us/j/93889521556?pwd=
eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1166
DTSTART;TZID=Asia/Kolkata:20211008T171500
DTEND;TZID=Asia/Kolkata:20211008T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1167
DTSTAMP:20230914T125953Z
SUMMARY:A search-to-decision reduction for minimizing formulas
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nThe Minimum Circuit Si
ze problem is a fundamental problem in theoretical computer science\, conn
ecting cryptography\, learning theory\, structural complexity\, etc.\, One
of the longstanding open problems is whether determining the size of a sm
allest circuit for a boolean function is equivalent to finding a minimum-s
ized circuit for the function. If one can find a minimum-sized circuit\, o
ne can certainly determine the minimum size required to compute a given fu
nction. The converse -- a search-to-decision reduction -- is not known. We
haven't yet ruled out the possibility that the decision problem takes lin
ear time but the search problem requires exponential time.\n\nIlango\, in
his CCC2020 paper\, makes progress in connecting the search and decision c
omplexity for minimizing formulas for any given boolean function. The main
result that we'll discuss today is that given an oracle to MFSP (Minimum
Formula Size Problem)\, one can solve Search-MFSP in time polynomial in th
e length N of the input (which is the truth table of the function f) and t
he number t of "near-optimal" formulas for f\, in time O(poly(N\,t)). Whil
e this quantity t is not well-understood\, we will see how this result giv
es us a deterministic O(2^{N/loglog(N)})-time oracle algorithm for solving
Search-MFSP on all but a o(1) fraction of instances.\n \nZoom Link: ht
tps://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1167
DTSTART;TZID=Asia/Kolkata:20211022T171500
DTEND;TZID=Asia/Kolkata:20211022T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1168
DTSTAMP:20230914T125953Z
SUMMARY:The Complexity of Simple Stochastic Games
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nA Simple Stochastic Game i
s a game with a reachability objective played by two players on a directed
graph. Each vertex of the graph is either controlled by one of the player
s or is a probabilistic vertex. The game begins by placing a token on the
start vertex. In each turn\, if the token is on a probabilistic vertex\, t
hen the token moves to one of its out-neighbours uniformly at random. Else
\, the player controlling the vertex chooses an out-neigbour to move the t
oken to. Player 1 wins if the token ever reaches the target vertex and Pla
yer 0 wins otherwise.\nGiven an SSG\, we would like to find out which play
er has a greater probability of winning the game. Condon (1992) showed tha
t this problem is in NP \\intersection coNP. There are some special cases
of the game for which the problem can be solved in polynomial time. In tod
ay's talk\, we discuss proofs of the same.\nLink to paper: https://doi.org
/10.1016/0890-5401(92)90048-K\nZoom Link: https://zoom.us/j/93889521556?
pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1168
DTSTART;TZID=Asia/Kolkata:20211029T171500
DTEND;TZID=Asia/Kolkata:20211029T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1169
DTSTAMP:20230914T125953Z
SUMMARY:Finding a satisfactory permutation
DESCRIPTION:Speaker: Kshitij Gajjar (National University of Singapore)\n\nA
bstract: \nYou have n candidates to fill up n vacant positions in an offic
e. The question is which candidate gets which position? To decide this\, y
ou ask non-candidates to vote. There are n! ways to allocate the positions
\, each corresponding to a permutation over [n]. Different voters may come
up with different permutations\, and your task is to find a permutation t
hat satisfies all the voters (or minimizes their dissatisfaction).\nThis p
roblem has several applications\, namely in computational biology\, DNA st
orage systems\, speech recognition\, social choice theory\, and classifica
tion. Surprisingly\, no poly(n)-time algorithm is known for this problem\,
even when there are only 4 voters. In this talk\, I will present a 1.5-ap
proximation algorithm for this problem\, whose runtime is poly(n) time whe
n the number of voters is a constant. This is joint work with Diptarka Cha
kraborty (NUS) and Agastya Vibhuti Jha (EPFL).\n\nZoom link: https://zoom.
us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1169
DTSTART;TZID=Asia/Kolkata:20211105T171500
DTEND;TZID=Asia/Kolkata:20211105T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1170
DTSTAMP:20230914T125953Z
SUMMARY:Upper bound on randomness complexity of private computations of AND
DESCRIPTION:Speaker: Hari Krishnan P A\n\nAbstract: \nIn a secure multi-par
ty computation problem\, players are required to compute a function of the
ir private inputs without revealing any extra information about this input
to other players. Randomness complexity is the number of random bits used
by the protocol which enables such a computation. It was previously known
that XOR can be computed using only one random bit for any number of play
ers.\n\nIn this talk\, we will see the result by Kushilevitz et al. which
shows that there exists a protocol that can privately compute the Boolean
function AND with 8 random bits for n>3 players and 7 bits for n=3 players
under a semi-honest adversarial setting.\n\nLink to the paper: https://ep
ubs.siam.org/doi/pdf/10.1137/20M1314197\n\nZoom link: https://zoom.us/j/9
3889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1170
DTSTART;TZID=Asia/Kolkata:20211112T171500
DTEND;TZID=Asia/Kolkata:20211112T181500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1171
DTSTAMP:20230914T125953Z
SUMMARY:Seroprevalence-based vaccination strategies: A combined transmissio
n and operational modeling approach.
DESCRIPTION:Speaker: Sarang Deo (Indian School of Business)\n\nAbstract: \n
Many low- and middle-income countries face limited supply of vaccines. In
such situations it is imperative to devise vaccination rollout strategies
that maximize the cost-effectiveness of these limited vaccine stocks. For
instance\, it may seem that selectively targeting regions with lower serop
revalence (e.g.\, rural areas) may be more effective in protecting more su
sceptible individuals than regions with higher seroprevalence (e.g.\, urba
n areas) that have fewer susceptible individuals. However\, these areas ma
y also have different reproduction rates which are the core drivers of dif
ferent seroprevalence. Similarly\, cost of supplying vaccines to these reg
ions may be different. As a result\, it is not obvious if\, and when\, suc
h seroprevalence driven strategies may be more effective than uniform vacc
ine rollout. We develop a modeling framework that combines transmission mo
deling with supply chain modeling to answer this question and illustrate i
ts utility using data on COVID-19 pandemic for the state of Punjab.\n\nYou
Tube link: https://www.youtube.com/watch?v=g3i1I-8lB1Q\n \n
URL:https://www.tcs.tifr.res.in/web/events/1171
DTSTART;TZID=Asia/Kolkata:20211116T160000
DTEND;TZID=Asia/Kolkata:20211116T170000
LOCATION:Via Zoom (please write to supriya.pottipati@tifr.res.in for zoom l
ink)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1172
DTSTAMP:20230914T125953Z
SUMMARY:On the Existence of EFX Allocations
DESCRIPTION:Speaker: Bhaskar Ray Chaudhury (University of Illinois at Urban
a-Champaign)\n\nAbstract: \nWe consider the problem of dividing indivisibl
e resources among a set of agents ``fairly''. Our underlying fairness noti
on is envy-freeness up to any good (EFX)\, where no agent envies another f
ollowing the removal of any single good from the other's bundle. Despite s
ubstantial effort from the community\, the existence of EFX allocations ha
s not been settled. In this talk\, we elaborate the proof of existence of
certain relaxations of EFX allocations and the existence of EFX allocatio
ns when there are only three agents. In the end\, we also reduce the probl
em of finding improved relaxations of EFX allocations to a problem in extr
emal graph theory.\n
URL:https://www.tcs.tifr.res.in/web/events/1172
DTSTART;TZID=Asia/Kolkata:20211123T173000
DTEND;TZID=Asia/Kolkata:20211123T183000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1173
DTSTAMP:20230914T125953Z
SUMMARY:Hardness and Independence of Polynomials
DESCRIPTION:Speaker: Prerona Chatterjee\n\nAbstract: \nAlgebraic Complexit
y Theory is a field in which one studies complexity theoretic questions su
rrounding algebraic objects. In this talk we will be broadly discussing tw
o such problems.\nThe first problem is showing lower bounds for explicit p
olynomials against various algebraic computational models. The most natura
l and well studied models of computation are algebraic circuits\, algebrai
c branching programs (ABPs) and algebraic formulas. With respect to provin
g lower bounds against these models\, we show the following results.\n1. A
ny algebraic branching program computing \\sum_{i=1}^n x_i^n must have at
least n^2 vertices. The previous best known lower bound was \\Omega(n log
n) on the number of edges for the same polynomial [Baur-Strassen].\n2. Any
formula computing the elementary symmetric polynomial of degree 0.1n must
have at least n^2 vertices. The previous best lower bound for any multili
near polynomial was \\Omega(n^2/log n) [Nechiporuk\, Kalorkoti]. It can al
so be shown that previous known methods can not prove a bound better than
\\Omega(n^2/log n) for any explicit multilinear polynomial.\nThis is joint
work with Mrinal Kumar\, Adrian She and Ben Lee Volk.\nAlong with proving
lower bounds against these models\, studying their relative powers is als
o an important problem in algebraic circuit complexity. It is known that f
ormulas can be efficiently simulated by ABPs and checking whether the conv
erse of this statement holds is a central question in the field. We make p
rogress towards solving this problem in the non-commutative setting\, wher
e we show a tight super-polynomial separation between ABPs and some struct
ured formulas.\nThe second problem that we are interested in relates the q
uestions of checking whether a given algebraic compuational model is compu
ting the zero polynomial or not and checking whether a given set of polyno
mials is algebraically independent or not. The connection between these qu
estions is via the notion of Faithful Homomorphisms. Although construction
of faithful homomorphisms were known when the underlying field had charac
teristic zero [Beecken-Mittman-Saxena\, Agrawal-Saha-Saptharishi-Saxena]\,
they were not known in the setting where the underlying field had finite
characteristic since efficient algorithms to check algebraic indepndence w
ere not known in this setting. Following up on the work of Pandey\, Saxena
and Sinhababu\, we construct faithful homomorphisms over fields of fini
te characetristics in some restricted settings and as a consequence show e
fficient polynomial identity tests for related models of computation. This
is joint work with Ramprasad Saptharishi.\n\nZoom link: https://zoom.us/j
/95369375923?pwd=VTlRNzNWUEdsbVRNb3RiSC9NdnRGZz09\n
URL:https://www.tcs.tifr.res.in/web/events/1173
DTSTART;TZID=Asia/Kolkata:20211126T173000
DTEND;TZID=Asia/Kolkata:20211126T183000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1174
DTSTAMP:20230914T125953Z
SUMMARY:Partial Function Extension with Applications to Property Testing an
d Learning
DESCRIPTION:Speaker: Gunjan Kumar\n\nAbstract: \nPartial function extension
is a basic problem that underpins multiple research topics in optimizatio
n\, including learning\, property testing\, and game theory. Here\, we are
given a partial function consisting of some points from a domain and a fu
nction value at each point. Our objective is to determine if this partial
function can be extended to a function defined on the domain\, that additi
onally satisfies a given property\, such as monotonicity. We formally stud
y partial function extension for various complement-free functions.\nOur c
ontributions are twofold. Firstly\, for the properties studied\, we give a
combinatorial characterization and bounds on the complexity of partial fu
nction extension. Secondly\, we develop new connections between partial fu
nction extension and learning and property testing\, and use these to give
new results for these problems.\nJoin Zoom Meeting\nhttps://zoom.us/j/967
20853429?pwd=ZUFFdzYvSmQzMWRNOG9PZVNzMHJpdz09\nMeeting ID: 967 2085 3429\n
Passcode: 869836\n
URL:https://www.tcs.tifr.res.in/web/events/1174
DTSTART;TZID=Asia/Kolkata:20211202T180000
DTEND;TZID=Asia/Kolkata:20211202T190000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1175
DTSTAMP:20230914T125953Z
SUMMARY:Online Bipartite Matching and Adwords
DESCRIPTION:Speaker: Vijay V. Vazirani (University of California\, Irvine)\
n\nAbstract: \nOver the last three decades\, the online bipartite matching
(OBM) problem has emerged as a central problem in the area of Online Algo
rithms. Perhaps even more important is its role in the area of Matching-Ba
sed Market Design. The resurgence of this area\, with the revolutions of t
he Internet and mobile computing\, has opened up novel\, path-breaking app
lications\, and OBM has emerged as its paradigmatic algorithmic problem.\n
In a 1990 joint paper with Richard Karp and Umesh Vazirani\, we gave an op
timal algorithm\, called RANKING\, for OBM\, achieving a competitive ratio
of (1 – 1/e)\; however\, its analysis was difficult to comprehend. Over
the years\, several researchers simplified the analysis.\nWe will start b
y presenting a “textbook quality” proof of RANKING. Its simplicity rai
ses the possibility of extending RANKING all the way to a generalization o
f OBM called the adwords problem. This problem is both notoriously difficu
lt and very significant\, the latter because of its role in the AdWords ma
rketplace of Google. We will show how far this endeavor has gone and what
remains. We will also provide a broad overview of the area of Matching-Bas
ed Market Design and pinpoint the role of OBM. \nBased on:\nhttps://arxiv
.org/pdf/2107.10777.pdf\nBio: Vijay Vazirani got his undergraduate degree
from MIT in 1979 and his PhD from the University of California\, Berkeley
in 1983. He is currently a Distinguished Professor at the University of Ca
lifornia\, Irvine.\nVazirani has made fundamental contributions to several
areas of the theory of algorithms\, including algorithmic matching theory
\, approximation algorithms and algorithmic game theory\, as well as to co
mplexity theory\, in which he established\, with Les Valiant\, the hardnes
s of unique solution instances of NP-complete problems. Over the last four
years\, he has been working on algorithms for matching markets. He is one
of the founders of algorithmic game theory.\nIn 2001 he published Approxi
mation Algorithms\, which is widely regarded as the definitive book on the
topic. In 2007\, he published the co-edited book Algorithmic Game Theory.
Another co-edited book\, Online and Matching-Based Market Design\, will b
e published by Cambridge University Press in early 2022\; see its flyer:
\nhttps://www.ics.uci.edu/~vazirani/flyer.pdf\n
URL:https://www.tcs.tifr.res.in/web/events/1175
DTSTART;TZID=Asia/Kolkata:20211207T160000
DTEND;TZID=Asia/Kolkata:20211207T170000
LOCATION:in person @ R.No. AG-69 and also via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1176
DTSTAMP:20230914T125953Z
SUMMARY:Secure Multiparty Computation with Limited Connectivity
DESCRIPTION:Speaker: Varun Narayanan\n\nAbstract: \nInformation theoretical
ly secure multiparty computation (MPC) is a central primitive in modern cr
yptography.\nIt enables mutually distrusting parties to collaboratively pe
rform computations on their combined data by ensuring that each party's da
ta is kept private from the others.\nThis is achieved by designing communi
cation protocols which allow the parties to collectively simulate an incor
ruptible trusted party\, who privately receives inputs from the parties\,
computes the pre-agreed functionality\, and delivers the outputs to the ap
propriate parties privately.\nThe subject of this dissertation is MPC when
there is limited connectivity in the communication network available to t
he participants.\nOur motivations and the progress we made in addressing t
hem follows:\n- In many practical scenarios\, the parties may only have ac
cess to a communication network with limited connectivity\, in that\, not
every pair of parties can communicate privately and reliably with each oth
er.\nWe characterize the conditions under which a pair of parties can comp
ute any functionality with information theoretic security in an incomplete
network of reliable\, private links.\nSeparate characterizations are obta
ined for honest-but-curious and malicious modes of corruption with securit
y against general adversary structures.\n- Many cryptographic tasks can be
modelled as secure 2-party computation (2PC) using only one-directional c
ommunication.\nGarg et al. (Crypto 15) initiated the study of non-interact
ive 2PC over noisy channels with one-way communication\, namely when only
one party speaks.\nA major question left open by that work was the complet
eness of finite channels in this model of secure computation.\nWe show tha
t bit-ROT (i.e.\, Randomized Oblivious Transfer) channel\, which erases on
e of the two input bits uniformly at random\, can compute any functionalit
y with inverse polynomial security error (in the number of channel uses) i
n this model against a computationally unbounded adversary.\nFurther\, ass
uming ideal obfuscation\, realizable using tamper-proof hardware tokens\,
naturally occurring channels such as binary symmetric channel (BSC) and bi
nary erasure channel (BEC) are complete in this sense with inverse polynom
ial security error against a computationally bounded adversary.\nTo comple
ment this\, we show that no channel with finite alphabet is complete in th
is model with negligible security error even against a computationally bou
nded adversary.\nFinally\, we characterize the channels that enable zero-k
nowledge proofs in this model\; the previous result work had presented con
struction of zero-knowledge proofs using BEC/BSC channels.\n- Studying sec
ure computation with limited interaction tends to reveal new frontiers to
approach the problem of complexity of several information theoretic primit
ives: a notoriously hard problem in cryptography.\nWe introduce a new prim
itive in information-theoretic cryptography\, namely zero-communication re
ductions (ZCR)\, with varying levels of security\, and relate it with many
other important primitives.\nUsing these connections\, we obtain new uppe
r bounds and lower bounds for complexity of these cryptographic primitives
.\n- MPC provides a meaningful and robust definition of security that can
be used for modelling security guarantees for existing models in network i
nformation theory.\nIndex coding is a well studied problem in which a serv
er wants to efficiently broadcast n messages intended for n users\, each w
ith access to a subset of these messages as side information.\nWe introduc
e a notion of privacy in index coding\, where the receivers do not learn a
nything more than the message they want from the server and those they hav
e as side information\, and study various aspects of its transmission rate
and secret consumption rate.\n\nJoin Zoom Meeting\nhttps://zoom.us/j/9507
4700581?pwd=dWVWZ0Y0WFVETGJYeDExRXJoN3BuQT09\nMeeting ID: 950 7470 0581\nP
asscode: 384118\n
URL:https://www.tcs.tifr.res.in/web/events/1176
DTSTART;TZID=Asia/Kolkata:20211207T173000
DTEND;TZID=Asia/Kolkata:20211207T183000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1177
DTSTAMP:20230914T125953Z
SUMMARY:Exact Sampling and List-Decoding
DESCRIPTION:Speaker: Siddharth Bhandari\n\nAbstract: \nIn this thesis defen
se\, we will discuss the following results.\n1. Exact sampling: We wil
l present an efficient algorithm that\, given a graph of maximum degree $\
\Delta$ and a list of at least $3\\Delta+1$ colours\, produces a random co
louring of the graph that is \\emph{exactly} uniformly distributed on the
set of all proper colourings. Before this work\, it was known that exact
sampling was possible if about $\\Delta^2$ colours were allowed.\n2. List
-decoding error-correcting codes: In list-decoding\, the decoder\, based
on the received word\, is required to output a small list of messages\, o
ne of which must be the original message. We will discuss the following re
sults about list-decoding.\n(a) Zero-error list decoding capacity of the $
q/(q-1)$ channel: We will present a lower bound showing that the zero-er
ror list decoding capacity of this channel is exponentially small in $q$ e
ven if the list size is allowed grow as $\\frac{1}{6} q \\ln q$. Previous
results showed that the capacity was exponentially small if the list size
was allowed to grow no larger than $1.58q$.\n(b) Multiplicity codes: We
consider a natural generalization of Reed-Muller codes where at each eva
luation point\, one records not only the evaluation of the message polyn
omial\, but also all its partial derivatives up to a certain order. We wil
l present an efficient algorithm that (under mild assumptions) list-deco
des multivariate multiplicity codes on arbitrary grids up to their distanc
e. Previously such results were known only for univariate multiplicity c
odes.\n(c) Polynomial ideal codes: A polynomial ideal code is specified
by a collection of relatively prime monic polynomials. The encoding is obt
ained by specifying the remainders of the message polynomial modulo the va
rious polynomials in the collection. We will describe an efficient algorit
hm to list-decode some special polynomial ideal codes\, which we \\emph{ca
ll affine folded Reed-Solomon codes\,} up to their distance. Previous resu
lts allowed list-decoding up to the distance for folded Reed-Solomon codes
\, univariate multiplicity codes and additive folded Reed-Solomon codes\,
which are all instances of affine folded Reed-Solomon codes.\n(The above r
esults were obtained in collaborations that involved the\nfollowing: Sayan
tan Chakraborty\, Prahladh Harsha\, Mrinal Kumar\, Jaikumar Radhakrishnan\
, Madhu Sudan.)\n\nZoom link: https://zoom.us/j/95587295066?pwd=ZW5QQ1NrSV
EwOG5iZzloVTlMV2ZBZz09\n
URL:https://www.tcs.tifr.res.in/web/events/1177
DTSTART;TZID=Asia/Kolkata:20211210T090000
DTEND;TZID=Asia/Kolkata:20211210T100000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1178
DTSTAMP:20230914T125953Z
SUMMARY:Convergence of nearest neighbor classification
DESCRIPTION:Speaker: Sanjoy Dasgupta (UC San Diego Jacobs School of Enginee
ring\nSan Diego\, California.)\n\nAbstract: \nThe "nearest neighbor (NN) c
lassifier" labels a new data instance by taking a majority vote over the k
most similar instances seen in the past. With an appropriate setting of k
\, it is capable of modeling arbitrary decision rules.\nTraditional conver
gence analysis for nearest neighbor\, as well as other nonparametric estim
ators\, has focused on two questions: (1) universal consistency---that is\
, convergence (as the number of data points goes to infinity) to the best-
possible classifier without any conditions on the data-generating distribu
tion---and (2) rates of convergence that are minimax-optimal\, assuming th
at data distribution lies within some standard class of smooth functions.\
nWe advance what is known on both these fronts. But we also show how to at
tain significantly stronger types of results: (3) rates of convergence tha
t are accurate for the specific data distribution\, rather than being gene
ric for a smoothness class\, and (4) rates that are accurate for the distr
ibution as well as the specific query point.\nAlong the way\, we introduce
a notion of "margin" for nearest neighbor classification. This is a funct
ion m(x) that assigns a positive real number to every point in the input s
pace\; and the size of the data set needed for NN (with adaptive choice of
k) to predict correctly at x is\, roughly\, 1/m(x).\nThe statistical back
ground needed for understanding these results is minimal\, and will be int
roduced during the talk.\nThis is joint work with Akshay Balsubramani\, Ka
malika Chaudhuri\, Yoav Freund\, and Shay Moran.\nBio:\nSanjoy Dasgupta wo
rks on unsupervised and minimally supervised learning. He is a professor o
f computer science at UC San Diego.\nYouTube livestream: https://www.youtu
be.com/watch?v=NMrPyN2neaw\n
URL:https://www.tcs.tifr.res.in/web/events/1178
DTSTART;TZID=Asia/Kolkata:20220111T103000
DTEND;TZID=Asia/Kolkata:20220111T113000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1179
DTSTAMP:20230914T125953Z
SUMMARY:Fourier Analytic Techniques in Machine Learning
DESCRIPTION:Speaker: Somnath Chakraborty\n\nAbstract: \nIn this talk\, we w
ill discuss the following two problems.\n \nProblem 1. Given independent r
andom samples drawn from an (almost uniform) mixture of at most k spherica
l d-dimensional Gaussians of unit variance\, can we devise an efficient al
gorithm that recovers the center's within arbitrary accuracy?\n\nProblem 2
. How can we integrate in the class of Lipschitz functions on d-dimensiona
l sphere\, computationally efficiently?\n\nEfficient algorithms are known
to deal with the first problem in the constant dimensions. We show that in
the regime d= log k\, Fourier analytic ideas can be employed to obtain an
efficient algorithm\, provided that minimum separation of the center's of
the individual Gaussian components is O(✓d). This minimum separation is
known to be optimal.\n\nFor the second problem\, we find a computationall
y efficient method of finding an equidistributed net on a d-dimensional sp
here. The techniques are inspired by work of Landau-Russell (2004) which u
sed ideas from Fourier analysis on finite groups to get a quantitative imp
rovement on the Alon-Roichman theorem that says a random Cayley graph is a
n expander.\n\nThese results are based on joint work with Hariharan Naraya
nan.\n\nJoin Zoom Meeting\nhttps://zoom.us/j/97251350514?pwd=UVA5aTR4eFlZT
VpER2dFRnZoRXJDUT09\n\nMeeting ID: 972 5135 0514\nPasscode: 844383\n
URL:https://www.tcs.tifr.res.in/web/events/1179
DTSTART;TZID=Asia/Kolkata:20220112T090000
DTEND;TZID=Asia/Kolkata:20220112T100000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1180
DTSTAMP:20230914T125953Z
SUMMARY:Stochastic window mean-payoff games
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nWe consider two-player sto
chastic games on a graph. Two-player stochastic games are a generalization
of two-player games and Markov decision processes. Each state in the gr
aph is controlled by one of the two players P1 and P2. The game begins by
placing a token on the initial state. In each turn\, the player controllin
g the state of the token chooses an action\, which then returns a probabil
ity distribution over the out-edges from the state. An out-edge is chosen
according to this distribution\, the token is moved along this edge to a n
ew state\, and the turn ends. The game then starts from the new state\, an
d the player who controls this new state chooses an action available from
that state.\nThis continues ad infinitum.\nIn this work\, we consider wind
ow mean-payoff objective. Each edge has a rational payoff. The sequence of
edges chosen in a play corresponds to a sequence of payoffs. Given an int
eger l\, and a threshold \\lambda\, the objective of player P1 is to ensur
e that from every state in a play\, for some interval window of length at
most l\, the mean of the payoffs in the window is at least \\lambda. The o
bjective of player P2 is the complement of P1's objective. Window mean-pay
off objectives have been studied for two-player games and for Markov decis
ion processes earlier. We study here two-player stochastic games with wind
ow objectives.\n\nZoom Meeting link:\nhttps://zoom.us/j/96361358353?pwd=U2
92aStIb3h5M1FIUTZWYThaL2dDdz09\nMeeting ID: 963 6135 8353\nPasscode: 70301
4\n
URL:https://www.tcs.tifr.res.in/web/events/1180
DTSTART;TZID=Asia/Kolkata:20220118T110000
DTEND;TZID=Asia/Kolkata:20220118T120000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1181
DTSTAMP:20230914T125953Z
SUMMARY:Designing Service Menus for Bipartite Queueing Systems
DESCRIPTION:Speaker: Varun Gupta (University of Chicago)\n\nAbstract: \nWe
look at the problem of designing a matching system with queues in an envir
onment with multiple types of customers and multiple classes of servers. C
ustomers arrive to the system seeking service by one of many available ser
vers\, and the reward or value obtained by a customer depends on both the
customer's type as well as the class of server which services the customer
. The goal of the system designer is to design a service mechanism that wi
ll match customers to servers and balance the two (usually) competing obje
ctives: (1) maximize customers' average service reward and (2) minimize cu
stomers' average waiting time delay. We restrict ourselves to a special cl
ass of mechanisms in which the system designer offers a static menu of ser
vice classes and customers choose which one of them to join upon arrival.
A service class is defined by a single queue served by a specific subset o
f servers under a First-Come-First-Served (FCFS) service discipline. Custo
mers act as rational self-interested utility maximizing agents when choosi
ng which service class to join. We study the problem under (conventional)
heavy traffic conditions\, that is\, in the limit as the traffic intensity
of the system approaches one from below\, and provide insights into the d
esign tradeoffs of "good" service menus.\nThe talk will be based on a work
-in-progress with Lisa Hillas and Rene Caldentey\, and the following paper
with Philipp Afeche and Rene Caldentey (https://papers.ssrn.com/sol3/pape
rs.cfm?abstract_id=3345302).\nBio: Varun Gupta studies stochastic modeling
and optimization\, applied probability\, algorithm design and analysis\,
and mechanism design. He is particularly interested in modeling and optimi
zation of resource allocation policies for multi-server and distributed sy
stems (e.g.\, third party logistics\, cloud infrastructure\, health care)
from a queueing theoretic perspective\, and learning and control in non-st
ationary environments.\nGupta holds a PhD in computer science from Carnegi
e Mellon University. He completed his undergraduate studies in computer sc
ience and engineering at the Indian Institute of Technology in Delhi where
he was awarded the President's Gold Medal. Outside of academia Varun has
spent time in industry at Alcatel-Lucent Bell Laboratories\, Microsoft Res
earch\, and Google Research.\n
URL:https://www.tcs.tifr.res.in/web/events/1181
DTSTART;TZID=Asia/Kolkata:20220118T183000
DTEND;TZID=Asia/Kolkata:20220118T193000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1182
DTSTAMP:20230921T105045Z
SUMMARY:A peek into Meta-Complexity
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nMeta-complexity refers
to the study of complexity of problems that are themselves about computat
ional complexity. The canonical meta-complexity problem in the Boolean wor
ld is MCSP (Minimum Circuit Size Problem). A recent result by Ilango has s
hone some light on when one can get an efficient Search-to-Decision reduct
ion for MFSP\, a variant of MCSP for formulas instead of circuits. We stud
ied their result in hopes of importing it to similar problems in the algeb
raic world. On the way\, we simplified a lemma from their paper and also o
bserved a surprising behaviour of random Boolean formulas\; these will be
the contents of this project seminar.\n
URL:https://www.tcs.tifr.res.in/web/events/1182
DTSTART;TZID=Asia/Kolkata:20220119T110000
DTEND;TZID=Asia/Kolkata:20220119T120000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1183
DTSTAMP:20230914T125953Z
SUMMARY:Lower bound on randomness complexity of secure computations of AND
DESCRIPTION:Speaker: Hari Krishnan P A\n\nAbstract: \nIn a secure multi-par
ty computation problem\, players are required to compute a function of the
ir private inputs without revealing any extra information about this input
to other players. Randomness complexity is the number of random bits used
by the protocol which enables such a computation. Since randomness is an
expensive resource\, developing MPC protocols to operate on minimum amount
of randomness has been an interesting problem. In a recent work\, Kushile
vitz et. al. prove a lower bound of 1 bit and an upper bound of 7 bits (8
bits) for the randomness complexity of securely computing Boolean AND func
tion in a 3-player (n>3 player) setting. In this work\, we improve upon th
is lower bound obtaining 3 bits for a 3-player setting using information t
heoretic tools.\nZoom link -\nhttps://zoom.us/j/98392241202?pwd=bTJRYklZZE
RXWWpIK0RIdnFZNzJSQT09\n
URL:https://www.tcs.tifr.res.in/web/events/1183
DTSTART;TZID=Asia/Kolkata:20220124T160000
DTEND;TZID=Asia/Kolkata:20220124T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1184
DTSTAMP:20230914T125953Z
SUMMARY:Sequential Hypothesis Testing under Stochastic Deadlines
DESCRIPTION:Speaker: Sushant Vijayan\n\nAbstract: \nPaper by: Peter Frazier
and Angela Yu (NIPS 2007).\nThe paper presents a generalisation to the cl
assic Wald-Wolfowitz result in the sequential hypothesis setting. They mod
ify the problem to incorporate stochastic deadlines and incorporate a more
general loss function. The paper solves the problem using dynamic program
ming techniques under a bayesian setting. I intend to talk a little bit ab
out the Wald-Wolfowitz solution and then discuss the paper.\n\nZoom link:\
nhttps://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1184
DTSTART;TZID=Asia/Kolkata:20220128T171500
DTEND;TZID=Asia/Kolkata:20220128T181500
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1185
DTSTAMP:20230914T125953Z
SUMMARY:A Study of Information Transmission over Quantum Channels in the On
e Shot Setting
DESCRIPTION:Speaker: Sayantan Chakraborty\n\nAbstract: \nWe study informat
ion transmission over quantum channels in the one shot setting. We primari
ly consider multiterminal channels and develop several tools to send quant
um information over these channels.\nOur first contribution is a generalis
ation of the well known decoupling theorem\, to the setting of multiple se
nders\, which immediately allows us to design encoders and decoders for en
tanglement transmission over the quantum multiple access channel\, among o
ther applications. However\, these results are non-smooth\, in the sense t
hat they do not recover the best known asymptotic iid bounds.\nTo overcome
this issue\, we develop two new techniques which we believe will be usefu
l in other contexts as well \;\ni. Successive cancellation decoding for en
tanglement transmission codes\nii. Quantum Rate Splitting\nFinally\, we ta
ckle the problem of sending private classical information over the quantum
MAC in the presence of an eavesdropper. Nothing was known about this prob
lem\, even in the asymptotic iid setting\, due to a famous open problem th
at becomes a bottleneck for most coding strategies. This problem is known
as 'simultaneous smoothing'. We overcome this issue by developing a new su
ccessive cancellation covering lemma\, which allows us to recover the idea
l rate region in the asymptotic iid setting. We also recover a non-trivial
rate region in the one shot setting by using our new covering lemma in co
njunction with our quantum rate splitting technique\n
URL:https://www.tcs.tifr.res.in/web/events/1185
DTSTART;TZID=Asia/Kolkata:20220201T140000
DTEND;TZID=Asia/Kolkata:20220201T150000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1186
DTSTAMP:20230914T125953Z
SUMMARY:Different flavours of differential privacy
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nDifferential privacy (DP)
gives a rigorous framework for data privacy by giving guarantees on the in
formation leakage for individual data points from the output of an algorit
hm. However\, DP has a somewhat paranoid view of the world which might be
too demanding for certain applications.\n\nIn this talk\, we shall first l
ook at some relaxations of DP - in particular R\\'enyi differential privac
y (RDP). One of the strengths of RDP is that it preserves the composition
properties of DP\, i.e. it is easy to bound the privacy loss of a sequence
of RDP mechanisms.\n\nThen\, we will do a quick overview of local differe
ntial privacy (LDP) - a notion of privacy useful for distributed applicati
ons. Time permitting\, we will also see how differential privacy is relate
d to generalization.\n\nZoom link:\nhttps://zoom.us/j/93889521556?pwd=eEFJ
WVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1186
DTSTART;TZID=Asia/Kolkata:20220204T160000
DTEND;TZID=Asia/Kolkata:20220204T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1187
DTSTAMP:20230914T125953Z
SUMMARY:Applying physics to mathematics
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nThis talk is based on Tada
shi Tokieda's wonderful talk (https://www.youtube.com/watchv=tQQ3oiB32GI&t
=1s) as a part of ICTS's distinguished lecture series. To quote Tokeida\,
"Humans tend to be better at physics than at mathematics. When an apple fa
lls from a tree\, there are more people who can catch it—we know physica
lly how the apple moves—than people who can compute its trajectory from
a differential equation. Applying physical ideas to discover and establish
mathematical results is therefore natural\, even if it has seldom been tr
ied in the history of science."\n\nWe will see how we can derive fundament
al theorems like Pythagoras theorem\, Cauchy-Schwarz inequality\, AM-GM in
equality among others using simple physics.\n\nZoom link:\nhttps://zoom.us
/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1187
DTSTART;TZID=Asia/Kolkata:20220211T160000
DTEND;TZID=Asia/Kolkata:20220211T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1188
DTSTAMP:20230914T125954Z
SUMMARY:Fano's method for lower bounds in machine learning
DESCRIPTION:Speaker: Neha Sangwan\n\nAbstract: \nWe will see how Fano's ine
quality\, a classical result from information theory\, can be used to get
tight lower bounds on sample complexity of various learning problems. We w
ill derive Fano's inequality and see its application on an example problem
.\n\nZoom link: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJG
QTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1188
DTSTART;TZID=Asia/Kolkata:20220218T160000
DTEND;TZID=Asia/Kolkata:20220218T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1189
DTSTAMP:20230914T125954Z
SUMMARY:Extension Preservation in the Finite and Prefix Classes of First Or
der Logic
DESCRIPTION:Speaker: Abhisekh Sankaran (University of Cambridge)\n\nAbstrac
t: \nIt is well known that the classic Los-Tarski preservation theorem fai
ls in the finite: there are first order definable classes of finite struct
ures closed under extensions which are not definable in the existential fr
agment of first order logic. We strengthen this [1] by constructing for ev
ery n\, first order definable classes of finite structures closed under ex
tensions which are not definable with n quantifier alternations. The class
es we construct are definable in the extension of Datalog with negation. T
his answers negatively an open question posed by Rosen and Weinstein [2].\
n[1] Anuj Dawar and Abhisekh Sankaran. Extension preservation in the finit
e and prefix classes of first order logic. Proceedings of the 31st Compute
r Science Logic (CSL)\, Ljubljana\, Slovenia\, January 25 -- 28\, 2021\, p
p. 18:1 -- 18:13.\n[2] Eric Rosen and Scott Weinstein. Preservation theore
ms in finite model theory. In the International Workshop on Logic and Comp
utational Complexity\, pages 480âEUR"502. Springer\,1994.\n\nBio: Abhisek
h Sankaran is a post-doctoral research associate at the Department of Comp
uter Science and Technology of the University of Cambridge\, UK. He works
with Prof. Anuj Dawar and has been at Cambridge since September 2018. Pri
or to that\, he was a post-doctoral fellow for a year in the Theoretical C
omputer Science Wing of the Institute of Mathematical Sciences Chennai. He
completed his Ph.D. and also his Bachelors and Masters from the Computer
Science department of IIT Bombay. His research interests lie in mathematic
al logic\, particularly classical and finite model theory\, and parameteri
zed algorithms\, particularly algorithmic metatheorems. For the present se
minar\, he will be speaking on a classical model theoretic result in the
context of finite structures.\n
URL:https://www.tcs.tifr.res.in/web/events/1189
DTSTART;TZID=Asia/Kolkata:20220222T160000
DTEND;TZID=Asia/Kolkata:20220222T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1190
DTSTAMP:20230914T125954Z
SUMMARY:Measurement Compression
DESCRIPTION:Speaker: Sayantan Chakraborty\n\nAbstract: \nOne of the fundam
ental components of quantum theory is measurement. one can think of measur
ement as the interface between the quantum world and our everyday world. A
n experimenter can gather information about a quantum state by doing measu
rements on it and use this information to design quantum protocols. Measur
ement\, however\, is a noisy process. The noise can arise from the measure
ment apparatus itself or from the uncertainty inherent in the quantum stat
e in this talk we will try to understand how one can separate out these tw
o types of noise. One can do this by formulating an information processing
task called measurement compression.\n\nMeasurement compression was first
proposed by Winter and subsequently generalised further by Wilde et al. H
owever the original proofs of this protocol are extremely complicated. The
talk will be based on new work by Chakraborty\, Padakandla and Sen by whi
ch we can obtain a much more elementary understanding of this fundamental
information processing task.\n\nZoom link: https://zoom.us/j/93889521556?p
wd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1190
DTSTART;TZID=Asia/Kolkata:20220225T160000
DTEND;TZID=Asia/Kolkata:20220225T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1191
DTSTAMP:20230914T125954Z
SUMMARY:Reconfiguring Shortest Paths in Graphs
DESCRIPTION:Speaker: Kshitij Gajjar (National University of Singapore)\n\nA
bstract: \nReconfiguring a system (generally represented by a graph) means
gradually transforming one configuration of the system to another configu
ration by modifying it in a slow and steady step-by-step manner. In the Sh
ortest Path Reconfiguration (SPR) problem\, we are given two shortest path
s\, and the goal is to transform one shortest path to the other by changin
g one vertex at a time\, so that all the intermediate configurations are a
lso shortest paths. SPR has many real-world applications like repaving roa
ds in a systematic way\, rerouting messages across a network of telecommun
ication towers\, cargo container stowage on ships\, train marshalling\, an
d rerouting data packets across servers in a synchronous multiprocessor se
tting.\n\nIn this talk\, we will show how one of these applications can be
modelled by a specific class of graphs\, and then we will solve SPR for t
hat graph class. We will also show that if we are allowed to change two ve
rtices at a time\, then SPR becomes PSPACE-complete\, even for a graph cla
ss for which SPR is known to be solvable in polynomial time when we are al
lowed to change only one vertex at a time. This is joint work with Agastya
Vibhuti Jha\, Manish Kumar and Abhiruk Lahiri (https://arxiv.org/abs/2112
.07499).\nZoom link: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmh
NYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1191
DTSTART;TZID=Asia/Kolkata:20220304T161500
DTEND;TZID=Asia/Kolkata:20220304T171500
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1192
DTSTAMP:20230914T125954Z
SUMMARY:Spectral Methods in Modern Graph Algorithms
DESCRIPTION:Speaker: Akash Kumar (EPFL\, Lausanne\, CH)\n\nAbstract: \nSpec
tral methods have had a strong influence on modern graph algorithms as evi
denced by the extensive literature on the subject. In my research so far\,
I have used spectral methods to develop algorithms for problems on planar
graphs\, and to develop algorithms to recover planted subgraphs in an oth
erwise gigantic random graph.\nIn this talk\, I will focus on my contribut
ions towards algorithmic problems on planar graphs. In particular\, I show
how random walk based (i.e.\, spectral) approaches led to progress on fin
ding forbidden minors [K.-Seshadhri-Stoman\, FOCS 2018] as well as on deci
ding planarity [K.-Seshadhri-Stolman\, STOC 2019] in bounded degree graphs
within the property testing framework. I will also cover how these approa
ches eventually led to progress on the so-called "efficient partition orac
le" problem [K.-Seshadhri-Stolman\, FOCS 2021].\nBio: Akash Kumar is a pos
tdoc at EPFL (Switzerland). His research interests lie in spectral graph t
heory and property testing. Before starting his postdoctoral position\, he
completed his PhD from Purdue University (USA) in May 2020 under the supe
rvision of Saugata Basu.\n\nJoin Zoom Meeting\nhttps://zoom.us/j/997106870
41?pwd=eHRvT1Y3OFJYMFk1QXNnVnFvMXZvUT09\nMeeting ID: 997 1068 7041\nPassco
de: 543679\n
URL:https://www.tcs.tifr.res.in/web/events/1192
DTSTART;TZID=Asia/Kolkata:20220308T160000
DTEND;TZID=Asia/Kolkata:20220308T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1193
DTSTAMP:20230914T125954Z
SUMMARY:Further Collapses in TFNP
DESCRIPTION:Speaker: Siddhartha Jain (EPFL\, Switzerland)\n\nAbstract: \nWe
show EOPL=PLS PPAD. Here the class EOPL consists of all total search prob
lems that reduce to the End-of-Potential-Line problem\, which was introduc
ed in the works by Hubacek and Yogev (SICOMP 2020) and Fearnley et al. (JC
SS 2020). In particular\, our result yields a new simpler proof of the bre
akthrough collapse CLS=PLS PPAD by Fearnley et al. (STOC 2021). We also pr
ove a companion result SOPL=PLS PPADS\, where SOPL is the class associated
with the Sink-of-Potential-Line problem.\n\nThe talk will be based on the
following paper: https://eccc.weizmann.ac.il/report/2022/018/\n\nZoom lin
k: https://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz09\n
URL:https://www.tcs.tifr.res.in/web/events/1193
DTSTART;TZID=Asia/Kolkata:20220311T161500
DTEND;TZID=Asia/Kolkata:20220311T171500
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1194
DTSTAMP:20230914T125954Z
SUMMARY:Hardness and Independence of Polynomials
DESCRIPTION:Speaker: Prerona Chatterjee\n\nAbstract: \nAlgebraic Complexit
y Theory is a field in which one studies complexity theoretic questions su
rrounding algebraic objects. In this talk we will be broadly discussing tw
o such problems.\nThe first problem is showing lower bounds for explicit p
olynomials against various algebraic computational models. The most natura
l and well studied models of computation are algebraic circuits\, algebrai
c branching programs (ABPs) and algebraic formulas. With respect to provin
g lower bounds against these models\, we show the following results.\n1. A
ny algebraic branching program computing \\sum_{i=1}^n x_i^n must have at
least n^2 vertices. The previous best known lower bound was \\Omega(n log
n) on the number of edges for the same polynomial [Baur-Strassen].\n2. Any
formula computing the elementary symmetric polynomial of degree 0.1n must
have at least n^2 vertices. The previous best lower bound for any multili
near polynomial was \\Omega(n^2/log n) [Nechiporuk\, Kalorkoti]. It can al
so be shown that previous known methods can not prove a bound better than
\\Omega(n^2/log n) for any explicit multilinear polynomial.\nThis is joint
work with Mrinal Kumar\, Adrian She and Ben Lee Volk.\nAlong with proving
lower bounds against these models\, studying their relative powers is als
o an important problem in algebraic circuit complexity. It is known that f
ormulas can be efficiently simulated by ABPs and checking whether the conv
erse of this statement holds is a central question in the field. We make p
rogress towards solving this problem in the non-commutative setting\, wher
e we show a tight super-polynomial separation between ABPs and some struct
ured formulas.\nThe second problem that we are interested in relates the q
uestions of checking whether a given algebraic compuational model is compu
ting the zero polynomial or not and checking whether a given set of polyno
mials is algebraically independent or not. The connection between these qu
estions is via the notion of Faithful Homomorphisms. Although construction
of faithful homomorphisms were known when the underlying field had charac
teristic zero [Beecken-Mittman-Saxena\, Agrawal-Saha-Saptharishi-Saxena]\,
they were not known in the setting where the underlying field had finite
characteristic since efficient algorithms to check algebraic indepndence w
ere not known in this setting. Following up on the work of Pandey\, Saxena
and Sinhababu\, we construct faithful homomorphisms over fields of fini
te characetristics in some restricted settings and as a consequence show e
fficient polynomial identity tests for related models of computation. This
is joint work with Ramprasad Saptharishi.\n
URL:https://www.tcs.tifr.res.in/web/events/1194
DTSTART;TZID=Asia/Kolkata:20220317T143000
DTEND;TZID=Asia/Kolkata:20220317T153000
LOCATION:in person @ R.No. AG-66 and also via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1195
DTSTAMP:20230914T125954Z
SUMMARY:Determinacy of Two-Player Games with Perfect Information
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nA game with *perfect infor
mation* is a game in which no information is hidden from the players. All
players know the rules and the state of the game at all times. Some exampl
es of games with perfect information are Chess\, Checkers\, Go\, and Nim.\
nIn two-player zero-sum games\, a player has a *winning strategy* if they
can always win no matter how the other player plays. A game is\n*determine
d* if one of the players has a winning strategy. Given a two-player game w
ith perfect information\, we would like to find out if it is determined.\n
In today's talk\, we discuss some results on two-player games with perfect
information:\n- We show that all finite games are determined.\n- We const
ruct an example of an infinite game that is not determined.\n- We define a
topology on infinite games. We use this to show the determinacy of some s
pecial classes of infinite games.\nAll results are from:\n[1] "Theory of G
ames and Economic Behavior" by von Neumann and Morgenstern\, 1947.\n[2] "I
nfinite Games with Perfect Information" by Gale and Stewart\, 1953.\n\nZoo
m link:\nhttps://zoom.us/j/93889521556?pwd=eEFJWVRtRHNpNlpZWmhNYTJGQTF6Zz0
9\n
URL:https://www.tcs.tifr.res.in/web/events/1195
DTSTART;TZID=Asia/Kolkata:20220325T161500
DTEND;TZID=Asia/Kolkata:20220325T171500
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1196
DTSTAMP:20230914T125954Z
SUMMARY:Context-Bounded Verification of Multithreaded Shared Memory Program
s
DESCRIPTION:Speaker: Ramanathan Thinniyam (Max Planck Institute for Softwar
e Systems)\n\nAbstract: \nMultithreaded shared memory programs appear in m
any applications such as operating systems\, web servers\, mobile applicat
ions etc. Verification of such programs has been of increasing concern sin
ce the early 2000s when clock speeds stabilised\, with the focus shifting
to architectures with multiple cores. From a theoretical perspective\, any
problem is undecidable already for programs with just two recursive threa
ds since one can then simulate a Turing machine. Hence we focus on restric
ting the problems to runs of the program which are context-bounded i.e. wh
ere every thread can be active at most K many times for some fixed number
K. This restriction has been effective at finding bugs since from a practi
cal standpoint most bugs occur already with a low number of context switch
es. We resolve long standing open problems related to the complexity of sa
fety and liveness verification of multithreaded programs in the presence o
f context bounding. This talk is aimed at giving a high level overview of
these results.\nBio: Ramanathan Thinniyam is currently a postdoc at the Ma
x Planck Institute for Software Systems in the Rigorous Software Engineeri
ng group. He obtained his BTech in Mechanical Engineering in IIT Madras\,
MSc in Theoretical Computer Science(TCS) at CMI and PhD in TCS at IMSc.\nH
e has been working on theoretical aspects of verification of multithreaded
programs and has published in conferences such as ICALP\, POPL and TACAS.
\n
URL:https://www.tcs.tifr.res.in/web/events/1196
DTSTART;TZID=Asia/Kolkata:20220405T160000
DTEND;TZID=Asia/Kolkata:20220405T170000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1197
DTSTAMP:20230914T125954Z
SUMMARY:Sketching and Streaming Complexity of Constraint Satisfaction Probl
ems
DESCRIPTION:Speaker: Madhu Sudan (Harvard John A. Paulson School of Enginee
ring and Applied Sciences)\n\nAbstract: \nIn this talk we will describe so
me of our recent work giving new upper and lower bounds on the approximabi
lity of constraint satisfaction problems (CSPs) in the streaming and sketc
hing settings. (Streaming algorithms process the input with small space\,
while sketching algorithms are restricted streaming algorithms that have a
dditional composability requirements.) When the sketching algorithms are c
onstrained to sub-polynomial space in the input length we get a fine dicho
tomy: CSPs on n variables are either solvable in polylogarithmic space or
require at least sqrt(n) space. Our positive results show the broad applic
ability of what we call ``bias-based sketching algorithms''\, and our nega
tive results work by abstracting and significantly generalizing previous b
ounds for the Maximum Cut problem. We will also mention some partial exten
sions to streaming algorithms\, linear space lower bounds\, ordering CSPs.
\nBased on joint work with Chi-Ning Chou\, Sasha Golovnev and Santhoshini
Velusamy.\n
URL:https://www.tcs.tifr.res.in/web/events/1197
DTSTART;TZID=Asia/Kolkata:20220408T100000
DTEND;TZID=Asia/Kolkata:20220408T110000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1198
DTSTAMP:20230914T125954Z
SUMMARY:An Algorithmic Approach to Uniform Lower Bound Conjectures
DESCRIPTION:Speaker: Rahul Santhanam (Department of Computer Science\nUnive
rsity of Oxford)\n\nAbstract: \nWe describe an algorithmic approach to uni
form lower bound conjectures such as NP != P and PSPACE != P.\n
URL:https://www.tcs.tifr.res.in/web/events/1198
DTSTART;TZID=Asia/Kolkata:20220408T160000
DTEND;TZID=Asia/Kolkata:20220408T170000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1199
DTSTAMP:20230914T125954Z
SUMMARY:Improved Quantum Query Upper Bounds Based on Classical Decision Tre
es
DESCRIPTION:Speaker: Nikhil S. Mande (CWI\, Amsterdam.)\n\nAbstract: \nIn t
he first part of this talk we will discuss the notion of rank of decision
trees\, which is essentially the largest depth of a complete subtree embed
ded in the initial tree. We observe the equivalence of rank and "guessing
complexity\," a measure of decision trees used by Lin and Lin [Theory of C
omputing'16] and Beigi and Taghavi [Quantum'20] to give upper bounds on qu
antum query complexity of functions based on classical query algorithms fo
r them.\n\nIn the second part of the talk we will first note that the best
speed-up obtainable using the approach of Beigi and Taghavi is captured b
y a polynomial optimization program on assignments of real weights to edge
s of the underlying classical decision tree. We then give a recursive expr
ession for the optimal solution to this program and bound the optimum from
above in terms of natural measures of the decision tree.\n\nBased on join
t work with Arjan Cornelissen and Subhasree Patro (https://arxiv.org/abs/2
203.02968).\n
URL:https://www.tcs.tifr.res.in/web/events/1199
DTSTART;TZID=Asia/Kolkata:20220426T160000
DTEND;TZID=Asia/Kolkata:20220426T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1200
DTSTAMP:20230914T125954Z
SUMMARY:Power-laws and weak convergence of the Kingman coalescent
DESCRIPTION:Speaker: Henrik Hult (KTH Royal Institute of Technology)\n\nAbs
tract: \nThe Kingman coalescent is an important and well-studied process i
n population genetics modelling the ancestry of a sample of individuals. I
n this talk weak convergence results are presented that characterise asymp
totic properties of the Kingman coalescent under parent dependent mutation
s\, as the sample size grows to infinity. It is shown that the sampling pr
obability satisfies a power-law and we derive the asymptotic behaviour of
transition probabilities of the block counting jump chain. For the normali
sed jump chain and number of mutations between types a limiting process is
derived consisting of a deterministic component\, describing the limit of
the block counting jump chain\, and independent Poisson processes with st
ate-dependent intensities\, exploding at the origin\, describing the limit
of the number of mutations. Finally\, the results are extended to charact
erise the asymptotic performance of popular importance sampling algorithms
\, such as the Griffiths-Tavare algorithm and the Stephens-Donnelly algori
thm. This is joint work with Martina Favero.\n
URL:https://www.tcs.tifr.res.in/web/events/1200
DTSTART;TZID=Asia/Kolkata:20220429T160000
DTEND;TZID=Asia/Kolkata:20220429T170000
LOCATION:A-201 Seminar Room
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1201
DTSTAMP:20230914T125954Z
SUMMARY:Shift\, scale and restart smaller models to estimate larger ones: A
gent based simulators in epidemiology
DESCRIPTION:Speaker: Daksh Mittal (TIFR Mumbai)\n\nAbstract: \nAgent-based
simulators are a popular epidemiological modelling tool to study the impac
t of various non-pharmaceutical interventions in managing an evolving pand
emic. They provide the flexibility to accurately model a heterogeneous pop
ulation with time and location varying\, person specific interactions. To
accurately model detailed behaviour\, typically each person is separately
modelled. This however\, may make computational time prohibitive when the
region population is large and when time horizons involved are large. In
this talk we review the agent based city simulator developed by IISC-TIFR
to model Covid epidemic and dig deeper into the underlying probabilistic
structure of the simulator (ABS) to arrive at modifications that allow sma
ller models to give accurate statistics for larger models. We exploit the
observations that in the initial disease spread phase\, the starting infec
tions create a family tree of infected individuals more-or-less independen
t of the other trees and are modelled well as a multi-type super-critical
branching process. Soon after\, once enough people have been infected\, th
e future evolution of the pandemic is closely approximated by its mean fie
ld limit with a random starting state. We build upon these insights to dev
elop a shift\, scale and restart algorithm for the simulator that accurate
ly evaluates the ABS's performance using a much smaller model. We provide
theoretical support for the proposed approach through an asymptotic analys
is where the population size increases to infinity.\n
URL:https://www.tcs.tifr.res.in/web/events/1201
DTSTART;TZID=Asia/Kolkata:20220506T163000
DTEND;TZID=Asia/Kolkata:20220506T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1202
DTSTAMP:20230914T125954Z
SUMMARY:A Simple Convergence Proof for A Simple Convergence Proof Stochasti
c Approximation and Applications to Reinforcement Learning
DESCRIPTION:Speaker: M. Vidyasagar (Indian Institute of Technology Hyderaba
d.)\n\nAbstract: \nSince its invention by Robbins and Monro in 1951\, the
stochastic approximation (SA) algorithm has been a widely used tool for fi
nding solutions of equations\, or minimizing functions\, with noisy measur
ements. Current methods for proving its convergence make use of the "ODE"
method whereby the sample paths of the algorithm are approximated by the t
rajectories of an associated ODE. This method requires a lot of technicali
ties. Interestingly\, as far back as 1965\, there was a paper by Gladyshev
that gave a simple convergence proof based on martingale methods\; howeve
r\, this proof worked for only a class of problems. In this talk I will co
mbine martingale methods with a new "converse theorem" for Lyapunov stabil
ity\, to arrive at a simple proof that works for the same situations where
the ODE method applies. The advantage of this approach is that it can pot
entially be applied to several problems in Reinforcement Learning (RL)\, s
uch as actor-critic learning (which is two time-scale SA)\, or RL with val
ue approximation (which is SA with projections onto a lower-dimensional su
bspace). These directions are under investigation.\nZoom Link - https://z
oom.us/j/91983281364?pwd=Wkl3MHMzWUFiYnVhV1d1U1E3bXhpZz09\n
URL:https://www.tcs.tifr.res.in/web/events/1202
DTSTART;TZID=Asia/Kolkata:20220510T160000
DTEND;TZID=Asia/Kolkata:20220510T170000
LOCATION:In person @ A-201 and also via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1203
DTSTAMP:20230914T125954Z
SUMMARY:Batch Asynchronous Stochastic Approximation and Applications to Tem
poral Difference Learning
DESCRIPTION:Speaker: M. Vidyasagar (Indian Institute of Technology Hyderaba
d.)\n\nAbstract: \n(This talk consists of joint work with Prof. Rajeeva L.
Karandikar) Reinforcement Learning (RL) algorithms such as Temporal Diffe
rence (TDL) or Q-learning update *just one component* of the value functio
n (TDL) or the action-value function (Q) at each time step. This is known
as asynchronous stochastic approximation. There are two issues with this.
First\, many of the "convergence proofs" in the literature are not always
correct. Second\, when the dimension of the state space is very high\, lea
rning requires a huge number of time steps. In effect\, spatial complexity
is replaced by temporal complexity. A compromise is to update *some but n
ot all* components of the value\, or the action-value\, function at each t
ime step. This may be called Batch Asynchronous Stochastic Approximation (
BASA). In this talk\, I will present a very general framework for proving
the convergence of BASA\, which includes both TD learning and Q-learning a
s special cases\, and also leads to new algorithms with lower complexity.\
nZoom Link - https://zoom.us/j/94084902566?pwd=KzNpb3NDQnorTWcvUlJvNXA3eH
lWUT09\n \n
URL:https://www.tcs.tifr.res.in/web/events/1203
DTSTART;TZID=Asia/Kolkata:20220511T110000
DTEND;TZID=Asia/Kolkata:20220511T120000
LOCATION:In person @ A-201 and also via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1204
DTSTAMP:20230914T125954Z
SUMMARY:Hilbert Schmidt Independence Criterion (HSIC)
DESCRIPTION:Speaker: Jatin Batra\n\nAbstract: \nSuppose we have access to $
n$ empirical observations of two random variables X\,Y and we want to know
- are X\,Y independent? One way to answer this is to compute empirical ve
rsions of various statistical quantities like covariance and mutual inform
ation. However\, because all we have access to is a finite number ($n$) of
empirical observations\, we might simply get unlucky. Can we guarantee th
at our test accuracy increases rapidly with $n$? The Hilbert Schmidt Indep
endence Criterion (HSIC) proposed by Gretton\, Bousquet\, Smola and Scholk
opft resolves this issue by providing an estimate of dependence that prova
bly gets more accurate at a $1/\\sqrt{n}$ rate. In this talk\, I will desc
ribe (following Gretton et al. in http://www.gatsby.ucl.ac.uk/~gretton/pap
ers/GreBouSmoSch05.pdf) how HSIC arises quite naturally as a kernel invari
ant version of the covariance estimate and also allude to some later appli
cations of HSIC.\n
URL:https://www.tcs.tifr.res.in/web/events/1204
DTSTART;TZID=Asia/Kolkata:20220513T160000
DTEND;TZID=Asia/Kolkata:20220513T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1205
DTSTAMP:20230914T125954Z
SUMMARY:Mixing time and expanders
DESCRIPTION:Speaker: Debsoumya Chakrabarti (IBS Discrete Mathematics Group)
\n\nAbstract: \nIn a graph\, we call a vertex well-mixing if the standard
random walk starting from that vertex reaches close to its stationary dist
ribution fast (say within 'polylogarithmic in the size of the vertex set'
steps). Call a graph well-mixing if all of its vertices are well-mixing. C
all a graph expander if every subset of vertices has a large boundary (in
other words\, the cut induced by every bipartition of the vertex set is 'l
arge'). For graphs\, it is well-known that the properties of being well-mi
xing and being an expander are equivalent (in a precise quantitative way).
Extending (in a weak sense) this classical result\, we establish that a r
egular graph is virtually an expander even if it has only a positive fract
ion of vertices that are well-mixing. Indeed\, we prove that such a graph
becomes an expander after deleting a small number of vertices. As a coroll
ary\, it shows that such graphs contain long cycles\, which improves a res
ult of Pak. Furthermore\, we can obtain such a cycle in polynomial time. T
his talk will be based on joint work with Jaehoon Kim\, Jinha Kim\, Minki
Kim\, and Hong Liu.\n\nYoutube link: https://youtu.be/N3XRrb3Bgzo\n
URL:https://www.tcs.tifr.res.in/web/events/1205
DTSTART;TZID=Asia/Kolkata:20220517T160000
DTEND;TZID=Asia/Kolkata:20220517T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1206
DTSTAMP:20230914T125954Z
SUMMARY:Radial projections in vector spaces over finite fields
DESCRIPTION:Speaker: Ben Lund (IBS Discrete Mathematics Group)\n\nAbstract:
\nSeveral recent papers by authors including Matilla\, Orponen\, Liu\, Sh
merikin\, and Wang give upper bounds on the Hausdorff dimension of the set
of points for which the radial projection of a Borel set in a real vector
space is much smaller than expected. In recent work\, joint with Thang Ph
am and Vu Thi Huong Thu\, we prove analogs of several of these theorems fo
r point sets in vector spaces over finite fields. In several cases\, we ar
e able to prove stronger bounds than the most natural analogs to the known
theorems in the continuous case. I will discuss these results\, and if ti
me permits I'll mention a connection to the Erdos and Falconer problems on
distinct distances.\n
URL:https://www.tcs.tifr.res.in/web/events/1206
DTSTART;TZID=Asia/Kolkata:20220520T160000
DTEND;TZID=Asia/Kolkata:20220520T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1207
DTSTAMP:20230914T125954Z
SUMMARY:Fast multivariate multipoint evaluation over finite fields
DESCRIPTION:Speaker: Mrinal Kumar (Indian Institute of Technology Bombay)\n
\nAbstract: \nMultipoint evaluation is the computational task of evaluatin
g a polynomial given as a list of coefficients at a given set of inputs. A
straightforward algorithm for this problem is to just iteratively evaluat
e the polynomial at each of the inputs. The question of obtaining faster-t
han-naive (and ideally\, close to linear time) algorithms for this problem
is a natural and fundamental question in computational algebra. In additi
on to its own inherent interest\, faster algorithms for multipoint evaluat
ion are closely related to fast algorithms for other natural algebraic que
stions like polynomial factorization and modular composition.\nNearly line
ar time algorithms have been known for the univariate multipoint evaluatio
n for close to five decades due to a work of Borodin and Moenck but fast a
lgorithms for the multivariate (or\, even bivariate) version have been muc
h harder to come by. In a significant improvement to the state of art for
this problem in 2008\, Umans and Kedlaya-Umans gave nearly linear time a
lgorithms for this problem over field of small characteristic and over all
finite fields respectively\, provided that the number of variables is at
most d^{o(1)} where d is the degree of the input polynomial in every varia
ble.\nIn this talk\, we will discuss two new algorithms for this problem:
the first is a simple and natural algebraic algorithm over not-too-large f
ields of small characteristic and the second is a (non-algebraic) algorith
m for this problem over all finite fields. Both these algorithms run in ne
arly linear time even when the number of variables is large. We will also
discuss an application to an upper bound for data structures for polynomia
l evaluation and to an upper bound on the rigidity of Vandermonde matrices
.\nThe talk is based on joint works with Vishwas Bhargava\, Sumanta Ghosh\
, Zeyu Guo\, Chandra Kanta Mohapatra and Chris Umans.\n
URL:https://www.tcs.tifr.res.in/web/events/1207
DTSTART;TZID=Asia/Kolkata:20220530T110000
DTEND;TZID=Asia/Kolkata:20220530T120000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1208
DTSTAMP:20230914T125954Z
SUMMARY:Data-derived weak universal consistency for lossless compression
DESCRIPTION:Speaker: Venkat Anantharam (University of California\, Berkeley
)\n\nAbstract: \nRich model classes for data may be too complex to admit u
niformly consistent estimators. In such cases\, it is conventional to sett
le for pointwise consistent estimators. But this viewpoint has the practic
al drawback that estimator performance is a function of the unknown model
within the model class that is being estimated. Even if an estimator is co
nsistent\, how well it is doing at any given time may not be clear\, no ma
tter what the sample size of the observations.\nWe explore how to resolve
this issue by studying model classes that may only admit pointwise consist
ency guarantees\, yet enough information about the unknown model driving t
he observations needed to gauge estimator accuracy can be inferred from th
e sample at hand. We would then say that such model classes admit data-der
ived weak universally consistent estimators.\nIn this work we flesh out th
is philosophy in the framework of lossless data compression problems over
a countable alphabet. Our main contribution is to characterize the model c
lasses that admit data-derived weak universally consistent lossless compre
ssion in terms of the presence or not of what we term deceptive distributi
ons (whether a distribution is deceptive or not is defined in the context
of the model class). We also show that the ability to estimate the redunda
ncy of compressing memoryless sources is equivalent to learning the underl
ying single-letter marginal in a data-derived fashion.\nThis is joint work
with Narayana Prasad Santhanam and Wojtek Szpankowski.\nBio: Venkat Anant
haram is on the faculty of the EECS department at U. C. Berkeley. He recei
ved his B. Tech. (1980) in Electrical Engineering (Electronics) from IIT M
adras\, and the M.S. (1982) and Phd. (1986) in Electrical Engineering and
M.A. (1984) and C. Phil. (1985) in Mathematics from U. C. Berkeley. From 1
986 to 1994 he was on the faculty of the School of EE at Cornell Universit
y. He has been with the EECS department at U. C. Berkeley since 1994. He i
s a Fellow of the IEEE and a Distinguished Alumnus of IIT Madras.\n
URL:https://www.tcs.tifr.res.in/web/events/1208
DTSTART;TZID=Asia/Kolkata:20220602T160000
DTEND;TZID=Asia/Kolkata:20220602T170000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1209
DTSTAMP:20230914T125954Z
SUMMARY:Techno-Economic Optimization Problems Related to 5G Technology
DESCRIPTION:Speaker: Gourav Saha (Purdue University)\n\nAbstract: \nSpectru
m sharing and Millimeter Wave Communication are two promising technologies
for 5G and beyond communication. In a spectrum sharing market\, the regul
ator (the government) and the wireless service providers are two important
stakeholders. In this talk\, I will focus mainly on spectrum sharing from
the perspective of the regulator.\nOne of the objectives of the regulator
is to maximize the utilization of the spectrum band. I will discuss a Sta
ckelberg game framework to optimize various parameters of the spectrum mar
ket in order to maximize spectrum utilization. These parameters include (i
) the duration of a spectrum license\, (ii) the number of spectrum bands\,
and (iii) the ratio of the licensed and unlicensed spectrum bands. Out of
these three parameters\, optimizing the duration of spectrum license is m
y most novel contribution and hence I will discuss this topic. Optimizing
the duration of spectrum license involves solving a Stackelberg game. I wi
ll discuss an O(log(T)) algorithm to solve the Stackelberg game\, T being
the maximum lease duration\, while the brute-force approach has a time com
plexity of O(T). I will also briefly discuss a combinatorial optimization
viewpoint of solving the Stackelberg Game.\nI will also briefly talk about
a variant of the Ski-Rental problem that we solved while addressing a cha
llenge faced by wireless service provider while operating in a spectrum sh
aring market. I will also briefly discuss my work on designing scheduling
algorithms for millimeter-wave communication by using tools from partially
observable Markov decision processes. Finally\, I will end the seminar wi
th future research plans related to (i) a variant of multi-armed bandits f
or directional millimeter-wave communication\, and (ii) spectrum enforceme
nt for spectrum sharing.\n\nBio: Gourav Saha received a B.E. degree from A
nna University\, Chennai\, India\, in 2012\, M.S. from Indian Institute of
Technology Madras\, India\, in 2015\, and Ph.D. from Rensselaer Polytechn
ic Institute\, Troy\, New York\, in 2020\, all in electrical engineering a
nd allied areas. He is currently a postdoctoral scholar in the Department
of Electrical and Computer Engineering of Purdue University and was previo
usly a postdoctoral scholar at Ohio State University. His research experie
nce includes control systems\, online algorithms\, game theory\, the econo
mics of wireless spectrum sharing market\, and Markov decision process. Hi
s current research involves designing scheduling and learning algorithms f
or millimeter-wave communication.\n
URL:https://www.tcs.tifr.res.in/web/events/1209
DTSTART;TZID=Asia/Kolkata:20220616T160000
DTEND;TZID=Asia/Kolkata:20220616T170000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1210
DTSTAMP:20230914T125954Z
SUMMARY:Secure Non-Interactive Reduction and Spectral Analysis of Correlati
ons
DESCRIPTION:Speaker: Varun Narayanan (Technion\, Israel.)\n\nAbstract: \nCo
rrelated pairs of random variables are a central concept in information-th
eoretically secure cryptography. Secure reductions between different corre
lations have been studied\, and completeness results are known. Further\,
the complexity of such reductions is intimately connected with circuit com
plexity and efficiency of locally decodable codes. As such\, making progre
ss on these complexity questions faces strong barriers. Motivated by this\
, in this work\, we study a restricted form of secure reductions --- namel
y\, Secure Non-Interactive Reductions (SNIR) --- which is still closely re
lated to the original problem\, and establish several fundamental results
and relevant techniques for it.\n\nWe uncover striking connections between
SNIR and linear algebraic properties of correlations. Specifically\, we d
efine the spectrum of a correlation\, and show that a target correlation h
as a SNIR to a source correlation only if the spectrum of the latter conta
ins the entire spectrum of the former. We also establish a `mirroring lemm
a' that shows an unexpected symmetry between the two parties in a SNIR\, w
hen viewed through the lens of spectral analysis. We also use cryptographi
c insights and elementary linear algebraic analysis to fully characterize
the role of common randomness as well as local randomness in SNIRs. We emp
loy these results to resolve several fundamental questions about SNIRs\, a
nd to define future directions.\n
URL:https://www.tcs.tifr.res.in/web/events/1210
DTSTART;TZID=Asia/Kolkata:20220624T113000
DTEND;TZID=Asia/Kolkata:20220624T123000
LOCATION:A-201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1211
DTSTAMP:20230914T125954Z
SUMMARY:Overcoming the sample complexity barrier in risk analytics with str
uctured tail models
DESCRIPTION:Speaker: Karthyek Murthy (Singapore University of Technology &
Design (SUTD))\n\nAbstract: \nThe ability to learn and control tail risks\
, besides being an integral part of quantitative risk management\, is impo
rtant in running operations requiring high service levels and cyberphysica
l systems requiring high reliability guarantees. Despite this significance
\, scalable algorithmic approaches have remained elusive due to the rarity
with which relevant risky samples are observed. In this talk we examine t
his bottleneck in two avenues: (i) statistical learning for minimization o
f tail risks and (ii) simulation of tail risks. We show efficient learning
and simulation is possible by exploiting the similarity with which risk e
vents unfold at different scales. This self-similarity\, being a nonparame
tric characteristic\, leads to satisfyingly expressive model classes and s
calable algorithms which require exponentially fewer samples than their be
nchmark counterparts. Efficient learning is made possible by a novel targe
ted approach towards robustness which could be of interest in broader cont
exts due to its automatic bias correction property. Specifically\, the sel
f-similar structure provides a fertile ground to exhibit how mildly restri
ctive structures can be utilized to debias the error introduced by first-s
tep model estimation.\n
URL:https://www.tcs.tifr.res.in/web/events/1211
DTSTART;TZID=Asia/Kolkata:20220628T160000
DTEND;TZID=Asia/Kolkata:20220628T170000
LOCATION:A-201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1212
DTSTAMP:20230914T125954Z
SUMMARY:STCS Vigyan Vidushi 2022
DESCRIPTION:Speaker: \n\nAbstract: \nSTCS Vigyan Vidushi 2022\, a summer sc
hool for women undergraduate and masters students will be held in TIFR Mum
bai campus from July 2 to July 15\, 2022.\nApplication deadline: April 25\
, 2022\n \n
URL:https://www.tcs.tifr.res.in/web/events/1212
DTSTART;VALUE=DATE:20220702
DTEND;VALUE=DATE:20220716
LOCATION:AG66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1213
DTSTAMP:20230914T125955Z
SUMMARY:Games of Pursuit and Evasion
DESCRIPTION:Speaker: Neeldhara Misra (IIT Gandhinagar)\n\nAbstract: \nPursu
it-Evasion games typically involve two types of players: those in pursuit
(aka cops) and the so-called evaders (aka robbers). The backdrop is usuall
y a network with specific rules that dictate how the players can move. The
se are turn-based games\, and one is usually interested in knowing whether
and how the evaders can be cornered. We often want to do this as inexpens
ively and quickly as we can\, which leads to questions about optimizing th
e number of cops we deploy and the number of rounds that the game will las
t. It turns out that answers to these questions often have deep connection
s with the structure of the underlying network. This talk will involve a f
ew glimpses of such connections. We will also discuss how these games can
model real-world situations.\n\nBio: Neeldhara Misra (BSc\, Mount Carmel C
ollege\, Bangalore\; PhD\, IMSc) is a Smt. Amba and Sri. V S Sastry Chair
Associate Professor at IIT Gandhinagar in the discipline of Computer Scien
ce and Engineering. Prior to this\, she was an INSPIRE faculty fellow at t
he department of Computer Science and Automation at the Indian Institute o
f Science\, and graduated from the Institute of Mathematical Sciences\, Ch
ennai. Her current research interests involve tackling computationally har
d problems with parameterized algorithms.\n
URL:https://www.tcs.tifr.res.in/web/events/1213
DTSTART;TZID=Asia/Kolkata:20220705T170000
DTEND;TZID=Asia/Kolkata:20220705T183000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1214
DTSTAMP:20230914T125955Z
SUMMARY:"Every Contact Leaves a Trace": Side Channel and Micro-architectur
al Attacks on Modern Cryptosystems
DESCRIPTION:Speaker: Prof. Debdeep Mukhopadhyay (Indian Institute of Techno
logy\nKharagpur)\n\nAbstract: \nPCryptography plays a vital role in securi
ng e-business and e-commerce transactions. However\, in spite of their mat
hematical robustness when these algorithms are implemented they may leak s
ensitive information via unintended side-channels. The focus of this talk
is to provide an overview of these side-channels which exist when the ciph
ers are executed on various platforms: from embedded systems to more gener
al-purpose computing systems. More precisely\, we start with introducing p
opular side-channels wherein the power consumption of a crypto-device can
leak the secret key. Subsequently\, we illustrate how faults can be exploi
ted. In the later part of the talk\, we illustrate how side-channels can a
lso be extended to more general-purpose computing systems. It is rather ir
onic that advancements in micro-architectures often are responsible for th
ese weaknesses. The talk attempts to elucidate the need of analyzing such
leakages to develop a holistic understanding of secured systems and transl
ate the theory to practice.\nAbout the Speaker: Prof. Debdeep Mukhopadhyay
is currently a Professor at the Department of CSE\, IIT Kharagpur. At IIT
Kharagpur he initiated Secured Embedded Architecture Laboratory (SEAL)\,
focusing on Hardware-Security. He had worked as\, visiting scientist at NT
U-Singapore\, visiting Associate Professor of NYU-Shanghai\, Assistant Pro
fessor at IIT Madras\, and Visiting Researcher at NYU-Tandon School of Eng
ineering\, USA. He holds a Ph.D\, M.S.\, and a B.Tech from IIT Kharagpur.
His research interests are on the topics of Cryptographic Engineering and
Hardware Security. Recently he is intrigued by adversarial attacks on mach
ine learning and encrypted computations.\nProf. Mukhopadhyay is the recipi
ent of the prestigious Shanti Swarup Bhatnagar Award 2021 for Science and
Technology and is a Fellow of the Indian National Academy of Engineers. He
was awarded Khosla National Award from IIT Roorkee 2021\, DST Swarnajayan
ti Fellowship 2015-16\, INSA Young Scientist award\, INAE Young Engineer A
ward\, and Associateship for the Indian Academy of Sciences.\n\nZoom link:
https://zoom.us/j/91427966752 (Passcode: 388629)\nYouTube live link: http
s://youtu.be/plvzy3fpY_E\n
URL:https://www.tcs.tifr.res.in/web/events/1214
DTSTART;TZID=Asia/Kolkata:20220708T160000
DTEND;TZID=Asia/Kolkata:20220708T170000
LOCATION:ONLINE ASET Colloquium
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1215
DTSTAMP:20230914T125955Z
SUMMARY:The Various Flavours of Computer Systems & Networking Research: A P
ersonal Journey
DESCRIPTION:Speaker: Ranjita Bhagwan (Microsoft Research Lab - India)\n\nAb
stract: \nSystems research encompasses a vast variety of domains\, problem
s\, and solution approaches. While in some cases the research problem is w
ell-defined\, in other cases defining the problem itself can be a major ta
sk. Sometimes we arrive at a solution through a novel and elegant idea tha
t strikes us suddenly\, while at other times we solve the problem by putti
ng together known ideas in interesting\, counter-intuitive ways. While som
e solutions are designed\, implemented\, and evaluated in a matter of week
s\, in other cases the whole process from conceptualization to realization
can take multiple years from start to finish. In this talk\, through prob
lems I have worked on over the years\, I will describe these various style
s of systems research and how each can be enjoyable and rewarding in its o
wn way.\n\nBio: Ranjita Bhagwan (BTech\, IIT Kharagpur\; MS and PhD\, Univ
California\, San Diego) is a Senior Principal Researcher at Microsoft Res
earch India. She has worked for more than a decade on applying machine lea
rning to improve system reliability\, security and performance. Recently\,
her work has focused on using data-driven approaches to improve cloud ser
vices and has led to severalpublications (including a best paper award at
USENIX OSDI 2018)\, as well as several tools that are widely used by Micro
soft's services. She is an ACM Distinguished Member and is the recipient o
f the 2020 ACM India Outstanding Contributions to Computing by a Woman Awa
rd.\n
URL:https://www.tcs.tifr.res.in/web/events/1215
DTSTART;TZID=Asia/Kolkata:20220711T170000
DTEND;TZID=Asia/Kolkata:20220711T183000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1216
DTSTAMP:20230914T125955Z
SUMMARY:One-way communication and non-adaptive decision trees
DESCRIPTION:Speaker: Swagato Sanyal (IIT Kharagpur)\n\nAbstract: \nFor two
Boolean functions f and g acting on n and m bits respectively\, their comp
osition f o g is a Boolean function on nm bits defined as follows. Its inp
ut is thought of as consisting of n blocks\, each m bits long. f o g is co
mputed first by computing g on each block\, and then by computing f on the
n resulting bits.\nThis talk is about one-way communication complexity of
composed functions. Here\, there are two communicating parties\, and the
input bits are distributed between them. One party sends a message to the
other\, based on which the other party outputs their guess of the value of
the function on the jointly held input.\nSuppose there is an algorithm fo
r f that queries few bits of f\, possibly randomly\, and outputs the value
of f. Suppose further that g is a function on very few bits. Then\, the t
wo communicating parties can simulate this algorithm and compute the compo
sed function by communicating about as many bits as the algorithm queries.
\nThe question we ask is if there is a communication protocol that is sign
ificantly cheaper than this naive protocol. We address this question for t
wo choices of g: the AND function and the Inner-Product function.\nThis ta
lk is based on joint work with Nikhil Mande and Suhail Sherif.\n\nLink to
the pre-print: https://arxiv.org/pdf/2105.01963.pdf\n
URL:https://www.tcs.tifr.res.in/web/events/1216
DTSTART;TZID=Asia/Kolkata:20220712T153000
DTEND;TZID=Asia/Kolkata:20220712T163000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1217
DTSTAMP:20230914T125955Z
SUMMARY:The complexity of formal proofs
DESCRIPTION:Speaker: Meena Mahajan (Institute of Mathematical Sciences\, Ch
ennai.)\n\nAbstract: \nA proof of a statement convinces the person/entity
addressed that the statement is true. Intuitively\, a good proof is short\
, and easy to verify. A formal proof must convince an automated checking p
rogram (that may have limited resources). This talk discusses why we care
about formal proofs\, how we can design good formal proofs\, and situation
s where we hit a wall. Bio: Meena Mahajan (BTech and MTech\, IIT Bombay\;
PhD\, IIT Madras) s a professor in the theoretical computer science group
at The Institute of Mathematical Sciences\, HBNI\, Chennai. Her research i
nterests encompass many aspects of computational complexity theory\, inclu
ding Boolean function complexity\, algebraic circuits\, proof complexity\,
and space-bounded computation. She is a fellow of the Indian Academy of S
ciences.\n
URL:https://www.tcs.tifr.res.in/web/events/1217
DTSTART;TZID=Asia/Kolkata:20220713T170000
DTEND;TZID=Asia/Kolkata:20220713T183000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1218
DTSTAMP:20230914T125955Z
SUMMARY:On Complexity measures of Boolean functions
DESCRIPTION:Speaker: Tulasi mohan Molli\n\nAbstract: \nBoolean functions ca
pture various problems and situations arising in computer science and othe
r areas. In this synopsis\, we study boolean functions using two complexit
y measures.\nIn the first part of the talk\, we will focus on the Probabil
istic degree of OR over Reals. This is based on joint work with Bhandari\,
Harsha and Srinivasan. In this part\, we will look at the construction of
a Probabilistic Polynomial for OR over Reals\, which improves on the prev
ious best construction due to Toda-Ogiwara and Beigel\, Tarui\, Reingold a
nd Speilman. We will also look at a lower bound on the Probabilistic deg
ree of OR which matches our upper bound construction in a restricted setti
ng.\n\nIn the second part\, we will look at a bunch of complexity measures
which arise out of the Fourier representation of Boolean functions and st
udy the relationship between them. This is based on joint work with Chak
raborty\, Mande\, Mittal\, Paraashar and Sanyal. In this part\, we will fo
cus on a couple of upper bounds on Fourier rank in terms of Fourier sparsi
ty\, weight\, Fourier max-entropy and Fourier max-rank entropy. We will
also exhibit functions which match these bounds.\n\nMeeting URL : https://
zoom.us/j/98513182619?pwd=WjZDWFhVVVdYM0NhZGZjdGtObjhvQT09\nMeeting ID : 9
8513182619\nPasscode : 59290580\n
URL:https://www.tcs.tifr.res.in/web/events/1218
DTSTART;TZID=Asia/Kolkata:20220715T140000
DTEND;TZID=Asia/Kolkata:20220715T150000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1219
DTSTAMP:20230914T125955Z
SUMMARY:Oral qualifier
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nA talk on
Pseudo-determinism.\n
URL:https://www.tcs.tifr.res.in/web/events/1219
DTSTART;TZID=Asia/Kolkata:20220718T103000
DTEND;TZID=Asia/Kolkata:20220718T113000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1220
DTSTAMP:20230914T125955Z
SUMMARY:Oral Qualifier
DESCRIPTION:Speaker: Yeshwant Chandrakant Pandit\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1220
DTSTART;TZID=Asia/Kolkata:20220719T150000
DTEND;TZID=Asia/Kolkata:20220719T163000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1221
DTSTAMP:20230914T125955Z
SUMMARY:Best arm Identification in Multi-Armed Bandits
DESCRIPTION:Speaker: Agniv Bandyopadhyay\n\nAbstract: \nPapers by: \n(1)"Be
st Arm Identification in Multi-Armed Bandits" by Audibert and Bubeck\, 201
0.\n(2)"Tight (Lower) Bounds for the Fixed Budget Best Arm IdentificationB
andit Problem" Carpentier and Locatelli 2016.\n\nPaper (1) gives two algor
ithms for the best arm identification problem in stochastic multi-armed ba
ndits under a fixed budget: Upper Confidence Bound-Exploration(UCB-E) and
Successive Reject(SR). They also prove a lower bound of error probability
in the limited budget setting. Paper (2) later improved that lower bound a
nd proved that it matches the upper bound of the probability of error for
the SR-Algorithm\, thus proving its optimality. We'll look at the results
established in Paper (1) and look through paper (2) if time permits.\n\nJo
ining Link: https://us02web.zoom.us/j/9290331190\n
URL:https://www.tcs.tifr.res.in/web/events/1221
DTSTART;TZID=Asia/Kolkata:20220720T150000
DTEND;TZID=Asia/Kolkata:20220720T160000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1222
DTSTAMP:20230914T125955Z
SUMMARY:Written Qualifier
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1222
DTSTART;TZID=Asia/Kolkata:20220721T093000
DTEND;TZID=Asia/Kolkata:20220721T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1223
DTSTAMP:20230914T125955Z
SUMMARY:Written Qualifier
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1223
DTSTART;TZID=Asia/Kolkata:20220722T093000
DTEND;TZID=Asia/Kolkata:20220722T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1224
DTSTAMP:20230914T125955Z
SUMMARY:An almost-efficient deterministic parallel algorithm for Bipartite
Perfect Matching
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nWe will try to underst
and the paper "Bipartite Perfect Matching is in quasi-NC" by Fenner\, Gurj
ar and Thierauf (2016). They achieved an almost complete derandomization o
f the Isolation Lemma (Mulmuley\, Vazirani\, Vazirani) for perfect matchin
gs in bipartite graphs and thus the result stated in the title. There are
no prerequisites for the talk\; familiarity with linear programming would
be helpful.\n
URL:https://www.tcs.tifr.res.in/web/events/1224
DTSTART;TZID=Asia/Kolkata:20220722T160000
DTEND;TZID=Asia/Kolkata:20220722T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1225
DTSTAMP:20230914T125955Z
SUMMARY:Learning Optimal Bids in Second Price Auctions with Temporal and Ov
erlapping Targeting Constraints
DESCRIPTION:Speaker: Ravi R. Mazumdar (University of Waterloo)\n\nAbstract:
\nAd placement in web-browsing and wireless mobiles is an increasingly im
portant component of the advertisement market. The market size is over $ 1
00 billion and counting. The mechanism is as follows: when a user opens a
webpage or mobile ap a message is sent to an exchange where multiple bidde
rs have the possibility of placing an ad that would target the right user\
, eg. age\, sex\, location\, etc. The ad that is displayed corresponds to
the bidder who bids the highest while the cost is calculated according to
a first or second price. Typically bidders are DSP (Demand Side Platforms)
that aggregate bids on behalf of clients who wish to run a campaign for a
given length of time with certain targeting criteria. The goal is to mini
mize the total cost of obtaining the required number of impressions (ads t
hat meet targeting criteria) over the duration of a contract. The real tim
e constraint is that bidding must be done within 100ms.\n\nIn this talk I
will build upon the theory that we had earlier developed for computing the
least cost bids in the second price context. This involves the notion of
an information state for the problem. There is a very rich primal-dual the
ory that emerges\, one in the so called impressions space and the other in
the contracts space. Computationally and structurally the primal and dual
views of the optimization have different properties that can be exploited
to come up with fast algorithms.\n\nThe optimal solutions depend on solvi
ng a constrained convex optimization problem when the information state is
known. However this is not readily available and thus there is the proble
m of learning the information state. We show that in the second price case
\, stochastic approximation (SA) algorithms that operate on censored data
(prices are only known by a bidder when the bidder wins) can be devised th
at solve the constrained optimization problem without learning the informa
tion state explicitly and we prove their convergence. Finally I will prese
nt the dynamic behaviour through simulations.\nJoint work with Ryan Kinnea
r (Waterloo) and Peter Marbach (Toronto). We thank Addictive Mobility Inc.
\, a Pelmorex company for having proposed the problem and to Addictive Mob
ility\, Ontario OCE VIP II\, and NSERC funding the work.\n\nBio: The speak
er was educated at the Indian Institute of Technology\, Bombay (B.Tech\, 1
977)\, Imperial College\, London (MSc\, DIC\, 1978) and obtained his PhD i
n Control Theory under A. V. Balakrishnan at UCLA in 1983. He is currently
a University Research Chair Professor in the Dept. of ECE at the Universi
ty of Waterloo\, Ont.\, Canada where he has been since September 2004. Pri
or to this he was Professor of ECE at Purdue University\, West Lafayette\,
USA. Since 2012 he is a D.J. Gandhi Distinguished Visiting Professor at t
he Indian Institute of Technology\, Bombay\, India and since May 2019 an A
djunct Professor at the Tata Institute of Fundamental Research (TIFR)\, Mu
mbai. He is a Fellow of the IEEE and the Royal Statistical Society. He is
a recipient of the INFOCOM 2006 Best Paper Award\, the ITC-27 2015 Best Pa
per Award\, the Performance 2015 Best Paper Award and was runner-up for th
e Best Paper Award at INFOCOM 1998. His research interests are in stochast
ic modelling and analysis applied to complex networks and statistical infe
rence.\nYouTube Link: https://youtu.be/dA44DA8gQDM\n
URL:https://www.tcs.tifr.res.in/web/events/1225
DTSTART;TZID=Asia/Kolkata:20220725T160000
DTEND;TZID=Asia/Kolkata:20220725T170000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1226
DTSTAMP:20230914T125955Z
SUMMARY:Oral Qualifier
DESCRIPTION:Speaker: Santanu Das\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1226
DTSTART;TZID=Asia/Kolkata:20220726T150000
DTEND;TZID=Asia/Kolkata:20220726T160000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1227
DTSTAMP:20230914T125955Z
SUMMARY:Oral Qualifier
DESCRIPTION:Speaker: Sunandan Sharma\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1227
DTSTART;TZID=Asia/Kolkata:20220728T150000
DTEND;TZID=Asia/Kolkata:20220728T160000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1228
DTSTAMP:20230914T125955Z
SUMMARY:Multi-Armed Bandits and Heavy-Tailed Distributions
DESCRIPTION:Speaker: Shubhada Agrawal\n\nAbstract: \nMulti-armed bandit (MA
B) is a popular framework for decision-making in an uncertain environment.
In its classical setup\, the algorithm has access to a fixed and finite s
et of unknown\, independent probability distributions or arms. At each tim
e step\, having observed the outcomes of all the previous actions\, the al
gorithm chooses one of the K arms and observes an independent sample drawn
from the underlying distribution\, which may be viewed as a reward. The a
lgorithm's goal is either to maximize the accumulated rewards or to identi
fy the "best" arm in as few samples as possible\, for an appropriate defin
ition of "best".\nVariants of these classical formulations have been widel
y studied. Tight lower bounds on associated performance metrics and algori
thms matching these lower bounds exactly have been developed assuming that
the arm-distributions are either Sub-Gaussian or come from a single param
eter exponential family (SPEF)\, for example\, Gaussian with known varianc
e or Bernoulli. However\, in practice\, the distributions may not belong t
o these simple classes. Developing lower bounds and optimal algorithms for
the general class of distributions largely remained open mainly because o
f the need for new sets of tools and techniques for the analysis in genera
lity.\nIn this dissertation\, we undertake a detailed study of the MAB pro
blems allowing for all the distributions with a known uniform bound on the
ir $(1+\\epsilon)^{th}$ moments\, for some $\\epsilon > 0$. This class s
ubsumes a large class of heavy-tailed distributions. We develop a framewor
k with essential tools and concentration inequalities and use it for desig
ning optimal algorithms for 3 different variants of the MAB problem. In th
is talk\, we will look at these optimal algorithms for the classical frame
works - regret minimization and best-arm identification.\nWe will also inv
estigate the new concentration inequalities that we develop for proving th
eoretical guarantees of the algorithms. These can be used to construct tig
ht\, anytime-valid confidence intervals (CIs) for various statistics\, for
example\, mean\, quantile\, CVaR\, etc.\, and may be of independent inter
est.\nThe above results were obtained in collaborations involving Sandeep
Juneja\, Wouter M. Koolen and Peter Glynn.\n\nZoom link: https://us02web.z
oom.us/j/9290331190\n
URL:https://www.tcs.tifr.res.in/web/events/1228
DTSTART;TZID=Asia/Kolkata:20220728T153000
DTEND;TZID=Asia/Kolkata:20220728T163000
LOCATION:Via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1229
DTSTAMP:20230914T125955Z
SUMMARY:Byzantine Multiple Access Channels
DESCRIPTION:Speaker: Neha Sangwan\n\nAbstract: \nWireless Internet of thing
s (IoTs) consists of devices with varying levels of security connected ove
r a shared communication medium. This can allow compromised devices to dev
iate from the communication protocol and potentially disrupt the communica
tion of other devices. The existing theoretical models of communication\,
for the most part\, assume users/devices which do not deviate maliciously
from the prescribed protocols. While attacks by external adversaries such
as jammers and eavesdroppers have been studied\, insider/ Byzantine attack
s have not received much attention. This is precisely the form that attack
s on IoTs take. As a first step towards this\, we consider the uplink of
a network and model it using a multiple access channel where users may de
viate maliciously. In this talk\, we will investigate this model under dif
ferent security guarantees: 1) When can honest devices communicate reliabl
y even in the presence of malicious devices? 2) Is it possible to identify
the malicious devices (when present) so that they can be taken offline? 3
) When is it possible to detect the presence of malicious devices (without
necessarily identifying them)?\n\nThe talk is based on joint work with Ma
yank Bakshi\, Bikash Kumar Dey and Vinod Prabhakaran.\n
URL:https://www.tcs.tifr.res.in/web/events/1229
DTSTART;TZID=Asia/Kolkata:20220729T140000
DTEND;TZID=Asia/Kolkata:20220729T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1230
DTSTAMP:20230914T125955Z
SUMMARY:Binary Hypothesis Testing with Deterministic Finite-Memory Decision
Rules
DESCRIPTION:Speaker: Hari Krishnan P A\n\nAbstract: \nIn this talk\, we wil
l consider the problem of binary hypothesis testing with finite memory. Co
nsider a sequence of IID random variables with expectation p under hypothe
sis H_0 and q under hypothesis H_1. Consider a finite-state machine with s
tate M_n at time n. Let the state of the system be governed by the rule M_
n = f(M_{n-1}\,X_n) where f is a deterministic time-invariant function. As
sume that we let this process run for a very long time and then make a dec
ision according to some mapping from the state space to the hypothesis spa
ce. We will bound the error probability of any hypothesis testing system.\
n
URL:https://www.tcs.tifr.res.in/web/events/1230
DTSTART;TZID=Asia/Kolkata:20220805T160000
DTEND;TZID=Asia/Kolkata:20220805T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1231
DTSTAMP:20230914T125955Z
SUMMARY:Gossiping in random graphs
DESCRIPTION:Speaker: Ayalvadi Ganesh (University of Bristol)\n\nAbstract: \
nConsider a set of n agents\, each of whom has a single message to convey
to all other agents. The messages are all of the same length. Time is divi
ded into rounds\, and during each round\, each agent may broadcast a singl
e message. Agents are represented as nodes of a directed communication gra
ph\, and a broadcast is received error-free by all (out)-neighbours of the
broadcasting node. The problem is to minimise the number of rounds until
all agents have received all messages.\n\nThis is known as the gossiping p
roblem\, and various versions of it have been studied. In our version\, th
e communication graph is a dense directed Erdos-Renyi random graph G(n\,p)
\, and we seek simple decentralised gossip algorithms. We consider two alg
orithms\, random relaying and random linear network coding. We consider a
sequence of graphs with p fixed and n tending to infinity. Our main result
s are that random relaying requires Theta(log n) rounds\, whereas random l
inear network coding requires only a constant number of rounds.\n
URL:https://www.tcs.tifr.res.in/web/events/1231
DTSTART;TZID=Asia/Kolkata:20220812T110000
DTEND;TZID=Asia/Kolkata:20220812T120000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1232
DTSTAMP:20230914T125955Z
SUMMARY:A game theoretic proof of RANKING algorithm
DESCRIPTION:Speaker: Soumyajit Pyne\n\nAbstract: \nOnline Bipartitie Matchi
ng was first introduced by Karp\, Vazirani and Vazirani (STOC’90). In th
eir seminal paper they had inroduced the RANKING algorithm which admits a
tight competitive ratio of 1-1/e. Since then multiple proofs of RANKING ha
ve been published. In this talk\, we shall look at a simple Game Theoretic
approach to proving the competitive ratio of RANKING\, avoiding linear pr
ogramming arguments. The proof is based on the paper by Eden\, Feldman\, F
iat and Segal (https://arxiv.org/abs/1804.06637).\n
URL:https://www.tcs.tifr.res.in/web/events/1232
DTSTART;TZID=Asia/Kolkata:20220812T160000
DTEND;TZID=Asia/Kolkata:20220812T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1233
DTSTAMP:20230914T125955Z
SUMMARY:Exact analysis of Generalization error in Generalized Linear Models
DESCRIPTION:Speaker: Parthe Pandit (UC San Diego)\n\nAbstract: \nGLMs are a
powerful class of models applied ubiquitously in machine learning and sig
nal processing applications. Learning these models often involves iterativ
ely solving non-convex optimization problems. I will present an exact stat
istical analysis of learning in these models in a high dimensional setting
. This framework is built on new developments in Random Matrix Theory\, an
d does not rely on convexity. Using this framework\, we can now analyze th
e effect of several design choices on the generalization error of the lear
ned model. Example design choices include loss function\, regularization\,
feature covariance\, train-test mismatch.\nBio: Parthe Pandit is a Simons
postdoctoral fellow at the Halıcıoğlu Data Science Institute at UC San
Diego. He obtained a PhD in ECE and MS in Statistics both from UCLA\, and
a dual degree in EE from IIT Bombay. He is a recipient of the Jack K Wolf
student paper award at ISIT 2019. He has also been a research intern at A
mazon and Citadel LLC.\n(Halıcıoğlu is pronounced as Haali-Jio-Lu)\n
URL:https://www.tcs.tifr.res.in/web/events/1233
DTSTART;TZID=Asia/Kolkata:20220816T160000
DTEND;TZID=Asia/Kolkata:20220816T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1234
DTSTAMP:20230914T125955Z
SUMMARY:KO codes: Inventing Non-linear Encoding and Decoding for Reliable W
ireless Communication via Deep-learning
DESCRIPTION:Speaker: Ashok Vardhan Makkuva (UIUC and EPFL)\n\nAbstract: \nL
andmark codes underpin reliable physical layer communication\, e.g.\, Reed
-Muller\, BCH\, Convolutional\, Turbo\, LDPC and Polar codes: each is a li
near code and represents a mathematical breakthrough. The impact on humani
ty is huge: each of these codes has been used in global wireless communica
tion standards (satellite\, WiFi\, cellular). Traditionally\, the design o
f codes has been driven by human-ingenuity and hence the progress is spora
dic. Can we automate and accelerate this process of discovering codes?\nIn
this talk\, I will talk about KO codes\, a new family of computationally
efficient deep-learning driven codes that outperform the state-of-the-art
reliability performance on the standardized AWGN channel. KO codes beat st
ate-of-the-art Reed-Muller and Polar codes\, under the low-complexity succ
essive cancellation decoding\, in the challenging short-to-medium block le
ngth regime on the AWGN channel. We show that the gains of KO codes are pr
imarily due to the nonlinear mapping of information bits directly to trans
mit real symbols (bypassing modulation) and yet possess an efficient\, hig
h performance decoder. The key technical innovation that renders this poss
ible is the design of a novel family of neural architectures inspired by t
he computation tree of the Kronecker Operation (KO) central to Reed-Muller
and Polar codes. These architectures pave the way for the discovery of a
much richer class of hitherto unexplored nonlinear algebraic structures. A
nd more interestingly\, despite having a lot of encoding and decoding stru
cture\, KO codes exhibit striking similarity to random Gaussian codes!\nBi
o: Ashok is an incoming postdoctoral associate at EPFL with Prof. Michael
Gastpar. He recently obtained his PhD in ECE from UIUC\, advised by Prof.
Pramod Viswanath. He also obtained his Masters in ECE (advised by Prof. Yi
hong Wu) from UIUC in 2017 and Bachelors in EE (advised by Prof. Vivek Bor
kar) with a minor in Mathematics from IIT Bombay in 2015. His current rese
arch interests are in theoretical and algorithmic aspects of machine learn
ing and information theory. He is a recipient of Best Paper Award from ACM
MobiHoc 2019. He is also a recipient of several graduate student awards a
nd fellowships including Joan and Lalit Bahl Fellowship (twice)\, Sundaram
Seshu International Student Fellowship\, and is a finalist for the Qualco
mm Innovation Fellowship 2018. Outside research\, he likes to learn new la
nguages\, watch and read about international films\, remembering movie tri
via and cooking. For more details about him\, please visit http://makkuva2
.web.engr.illinois.edu/\n
URL:https://www.tcs.tifr.res.in/web/events/1234
DTSTART;TZID=Asia/Kolkata:20220823T160000
DTEND;TZID=Asia/Kolkata:20220823T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1235
DTSTAMP:20230914T125955Z
SUMMARY:New Techniques and Results in Mixture Models
DESCRIPTION:Speaker: Soumyabrata Pal (Google Research\, India.)\n\nAbstract
: \nMixture models\, introduced in 1894 by Karl Pearson is very popular in
both theory and practice. Mixture models with high dimensional latent par
ameter vectors are widely used to fit complex multimodal datasets as they
allow representation of the presence of sub-populations within the overall
population. The primary difficulty in learning mixture models is that the
observed data does not identify the subpopulation to which an individual
observation belongs.\nIn this talk\, I will introduce the problem of suppo
rt recovery of the unknown parameter vectors when they are known to be spa
rse. I will present a very generic framework (including a novel tensor-bas
ed algorithm) for support recovery by using estimates of the number of unk
nown vectors having non-zero entries in small subsets of indices. Then\, t
his framework is applied by showing a variety of techniques to estimate th
e aforementioned quantities in different mixture models. Our results for s
upport recovery are quite general\, namely they are applicable to 1) Mixtu
res of many different canonical distributions 2) Mixtures of linear regres
sions and linear classifiers. Finally\, I will demonstrate some experiment
s on real world datasets that support our theoretical guarantees.\nBased o
n papers that appeared in NeurIPS 20\, NeurIPS 21 and AISTATS 22.\nBio: So
umyabrata is a Postdoctoral Researcher at Google Research\, India. Soumyab
rata completed his Ph.D in the Computer Science Department (CICS) at the U
niversity of Massachusetts Amherst advised by Dr. Arya Mazumdar. During th
at time\, he was a Visiting Graduate Student at the University of Californ
ia San Diego in 2021 and interned at Ernst & Young AI Lab and Amazon AI an
d Search. His research interests are Theoretical Machine Learning\, Applie
d Statistics and Information Theory. In particular\, his research is focus
ed on designing efficient and scalable algorithms for Statistical recovery
/reconstruction problems under different structural assumptions on the dat
a generating mechanism such as sparsity\, low-rank\, presence of latent cl
usters among others.\nWebsite: https://soumyabratap.github.io/\n
URL:https://www.tcs.tifr.res.in/web/events/1235
DTSTART;TZID=Asia/Kolkata:20220830T160000
DTEND;TZID=Asia/Kolkata:20220830T170000
LOCATION:Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1236
DTSTAMP:20230914T125955Z
SUMMARY:A Study of Information Transmission over Quantum Channels in the On
e-Shot Setting
DESCRIPTION:Speaker: Sayantan Chakraborty\n\nAbstract: \nIn this thesis we
tackle two main problems :\n1. Coding techniques for sending quantum info
rmation across a multi-terminal quantum channel in the one-shot regime.\n2
. Coding techniques for sending private information across a classical-qua
ntum multiple access channel\, while ensuring that the transmitted message
s remain hidden from an eavesdropper.\nWe first consider the Quantum Multi
ple Access Channel or QMAC and show the existence of coding theorems for t
he task of sending EPR pairs from two independent senders Alice and Bob to
a receiver Charlie using this channel. For this purpose we develop two ma
in tools:\n1. A multi-terminal decoupling theorem.\n2. One-shot quantum ra
te splitting.\nWhile the multi-terminal decoupling theorem allows us to re
cover an ideal pentagonal achievable rate region\, for entanglement transm
ission across the QMAC\, these bounds are not easily generalised to recove
r the best inner bounds known for this problem in the asymptotic iid setti
ng\, due to the issue of Simultaneous Smoothing\, which remains a major op
en problem in quantum information theory.\nTo get around this problem\, we
adapt the classical iid technique of rate splitting to the one-shot quant
um setting. This new technique allows us to recover the best known achieva
ble rate region known for this problem in the asymptotic iid regime. Our t
echniques also allow us to show the existence of coding schemes which achi
eve the best known non-trivial rate region for entanglement transmission a
cross the Quantum Interference Channel or QIC\, both in the one-shot and i
n the asymptotic iid regime.\nWe next consider the problem of sending priv
ate classical information from two independent senders Alice and Bob to a
legitimate receiver Charlie\, in the presence of an eavesdropper Eve\, via
a wiretap classical-quantum MAC. This problem is considerably hard to sol
ve since we require that Alice and Bob's messages should be jointly secret
from Eve\, which requires a joint quantum covering lemma.\nUnfortunately\
, such a lemma remains out of reach of current techniques\, since it reduc
es to the Simultaneous Smoothing conjecture.\nWe get around this issue by
developing a successive cancellation style covering lemma which\, along wi
th some other tools\, allows us to recover a non-trivial achievable rate r
egion in the one-shot regime and also recover the natural pentagonal rate
region that one would expect for this problem\, in the asymptotic iid regi
me.\n
URL:https://www.tcs.tifr.res.in/web/events/1236
DTSTART;TZID=Asia/Kolkata:20220901T173000
DTEND;TZID=Asia/Kolkata:20220901T193000
LOCATION:Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1237
DTSTAMP:20230914T125956Z
SUMMARY:A brief introduction to polar codes
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nIn his se
minal work in 1948 (A Mathematical Theory of Communication) Shannon introd
uced the capacity of a channel: the maximum rate at which information can
be reliably communicated across a channel. Shannon proved that for every m
emoryless channel there exists error correcting codes which achieve rate a
rbitrarily close to capacity. However Shannon's proof was an existential o
ne: he was unable to give any efficient encoding and decoding scheme that
would allow communication over a channel at rates close to capacity. This
remained a major open problem until 2008 when Erikan introduced polar code
s. These codes achieve Shannon capacity over any symmetric memoryless chan
nel\, and in special cases such as BSC (binary symmetric channel) and BEC
(binary erasure channel) have efficient decoding algorithms as well. We sh
all discuss polar codes in the special case of BSC: we shall give an overv
iew of their encoding and decoding schemes.\n
URL:https://www.tcs.tifr.res.in/web/events/1237
DTSTART;TZID=Asia/Kolkata:20220902T140000
DTEND;TZID=Asia/Kolkata:20220902T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1238
DTSTAMP:20230914T125956Z
SUMMARY:M.Sc. thesis waiver presentation
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nOne of th
e most active research directions in quantum computing currently is establ
ishing quantum supremacy results: coming up with computational tasks which
a quantum algorithm can solve super-polynomially faster than a classical
algorithm. In this talk we look at the particular computational model of q
uery complexity: where the algorithm has to evaluate a known function on a
n unknown input\; it can access the input by asking queries to an oracle a
nd has to minimize the number of queries. Internal computations are assume
d to have no cost. We can define the quantum variant of query complexity w
here the quantum algorithm has access to an oracle representing the input
and can ask queries in superposition.\n\nOne of the long-standing question
s in this direction is the following: given a quantum query algorithm for
a Boolean function (defined on a subset of the Boolean hypercube with at l
east inverse polynomial density) making d queries\, does there exist a cla
ssical query algorithm which makes at most poly(d) queries and approximate
s the output probability of the quantum algorithm on most inputs? We shall
discuss the body of previous work done on this question: weak results tha
t are known\, a recent attempt to resolve the conjecture which turned out
to have a fatal flaw and our attempts to fix the argument.\n
URL:https://www.tcs.tifr.res.in/web/events/1238
DTSTART;TZID=Asia/Kolkata:20220908T103000
DTEND;TZID=Asia/Kolkata:20220908T120000
LOCATION:D-405 (D-Block Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1239
DTSTAMP:20230914T125956Z
SUMMARY:YouTube video by Ryan O'Donnell
DESCRIPTION:Speaker: \n\nAbstract: \nPlease find the details of this week's
student seminar.\n\nThis week\, we shall be watching a YouTube video by R
yan O'Donnell. Here\, he will be presenting a complete proof of Hastad's S
witching Lemma. This video was suggested to me by Ashutosh.\n\nLink : http
s://www.youtube.com/watch?v=ahW96yYmWx0\n\nRegards\,\nArghya\n
URL:https://www.tcs.tifr.res.in/web/events/1239
DTSTART;TZID=Asia/Kolkata:20220909T160000
DTEND;TZID=Asia/Kolkata:20220909T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1240
DTSTAMP:20230914T125956Z
SUMMARY:The L* Algorihthm for Constructing Deterministic Finite Automatons
DESCRIPTION:Speaker: Malhar Ajit Managoli\n\nAbstract: \nSuppose we are pr
esented with a language like L = {x ∈ {0\, 1}| The third to last bit is
1}. We can intuitively understand this language. Given a string we can qui
ckly tell if it is in L or not. Coming up with a DFA for a language from s
cratch is somewhat more difficult.\n\nThe L* algorithm is one way of writi
ng a formal method which we use in finding a DFA for a regular language. I
n this talk we will see a description of the L* Algorithm and a proof of c
orrectness.\n
URL:https://www.tcs.tifr.res.in/web/events/1240
DTSTART;TZID=Asia/Kolkata:20220916T160000
DTEND;TZID=Asia/Kolkata:20220916T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1241
DTSTAMP:20230914T125956Z
SUMMARY:Binary linear error correcting codes with sufficient symmetry achie
ve capacity on the erasure channel
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nConstruct
ing error correcting codes that achieve Shannon capacity on a given memory
less channel and have efficient encoding and decoding algorithms has been
a major line of recent research. In this talk\, we shall discuss the error
correcting capacity of codes on the binary erasure channel. We will show
that sufficient symmetry alone implies a linear error correcting code achi
eves capacity on BEC. In particular\, we shall prove the following results
: (i) Consider a family of linear binary error correcting codes (parameter
ized by block length) approaching constant rate. Suppose the transitive gr
oup of each code is doubly symmetric and the distance satisfies log(distan
ce)/log(block length)->1. Then this family achieves capacity on BEC. This
implies that BCH codes achieve capacity on BEC. (ii) Binary Reed Muller co
des achieve capacity on BEC. In this proof\, we shall not use any special
properties of Reed Muller codes other than the fact that their symmetry gr
oup contains GL_n(F_2) and that a family of Reed Muller codes approaching
constant rate satisfies log(distance)/log(block length)->1/2.\n
URL:https://www.tcs.tifr.res.in/web/events/1241
DTSTART;TZID=Asia/Kolkata:20220923T160000
DTEND;TZID=Asia/Kolkata:20220923T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1242
DTSTAMP:20230914T125956Z
SUMMARY:Searching valiantly for LTCs
DESCRIPTION:Speaker: Ashutosh Shankar (TIFR\, Mumbai.)\n\nAbstract: \nA des
irable property for a code is "local testability" - being able to look at
a small number of locations and determine if the received word is a codewo
rd or far from one. Given two base codes\, one can construct a tensor code
: matrices with rows in one code\, and columns in the other. A natural loc
al test for this would be to randomly pick a row or column and check if it
is in that base code. Unfortunately\, this fails - we will look at a neat
counterexample by Paul Valiant (2005). However\, if one of the codes is "
smooth"\, the test works. Time permitting\, we will look at the definition
of smoothness (later relaxed to "weakly smooth") and a sketch of the anal
ysis in that case.\n
URL:https://www.tcs.tifr.res.in/web/events/1242
DTSTART;TZID=Asia/Kolkata:20220930T160000
DTEND;TZID=Asia/Kolkata:20220930T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1243
DTSTAMP:20230914T125956Z
SUMMARY:Universal Caching
DESCRIPTION:Speaker: Ativ Joshi (CMI)\n\nAbstract: \nIn the learning litera
ture\, the performance of an online policy is commonly measured in terms o
f the static regret metric\, which compares the cumulative loss of an onli
ne policy to that of an optimal benchmark in hindsight. In the definition
of static regret\, the benchmark policy remains fixed throughout the time
horizon. Naturally\, the resulting regret bounds become loose in non-stati
onary settings\, where fixed benchmarks often suffer from poor performance
. In this paper\, we investigate a stronger notion of regret minimization
in the context of an online caching problem. In particular\, we allow the
action of the offline benchmark at any round to be decided by a finite sta
te predictor possessing arbitrarily many states. Using ideas from the univ
ersal prediction literature in information theory\, we propose an efficien
t online caching policy with an adaptive sublinear regret bound. To the be
st of our knowledge\, this is the first data-dependent regret bound known
for the universal caching problem. We establish this result by combining a
recently-proposed online caching policy with an incremental parsing algor
ithm\, e.g. Lempel-Ziv '78. Our methods also yield a simpler learning-theo
retic proof of the improved regret bound\, as opposed to the more involved
and problem-specific combinatorial arguments used in the earlier works.\n
URL:https://www.tcs.tifr.res.in/web/events/1243
DTSTART;TZID=Asia/Kolkata:20221007T160000
DTEND;TZID=Asia/Kolkata:20221007T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1244
DTSTAMP:20230914T125956Z
SUMMARY:Streaming Algorithms for Maximum Weighted Matching
DESCRIPTION:Speaker: Pavel Dvorak\n\nAbstract: \nI will talk about the maxi
mum weighted matching problem in the streaming model. In this model\, an a
lgorithm receives edges of a graph one by one. When the stream ends\, it s
hould output an approximation of the maximum weighted matching of the inpu
t graph using only linear (or subquadratic) memory. First\, I will present
an algorithm of Paz and Swartzmann\, who gave a 2-approximation for this
problem. Further\, I will talk about how to use this algorithm for design
algorithms in sliding window model. In this model\, when an edge arrives\,
the algorithm should output an approximation of the maximum weighted matc
hing in a graph spanned by last L edges\, where L is a parameter of the mo
del.\n\nThis is a joint work with Cezar-Mihail Alexandru\, Christian Konra
d\, and Kheeran K. Naidu.\n
URL:https://www.tcs.tifr.res.in/web/events/1244
DTSTART;TZID=Asia/Kolkata:20221014T160000
DTEND;TZID=Asia/Kolkata:20221014T170000
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1245
DTSTAMP:20230914T125956Z
SUMMARY:Polynomial time partition oracles for bounded degree planar graphs
DESCRIPTION:Speaker: Akash Kumar (EPFL Lausanne\, Switzerland.)\n\nAbstract
: \nTake a planar graph with maximum degree d. These graphs admit a hyperf
inite decompositions\, where\, for a sufficiently small \\epsilon > 0\, on
e removes \\epsilon dn edges to get connected components of size independe
nt of n. An important tool for sublinear algorithms and property testing f
or such classes is the partition oracle. A partition oracle is a local pro
cedure that gives "consistent" access to a hyperfinite decomposition\, wit
hout any preprocessing. Given a query vertex v\, the partition oracle outp
uts the component containing v in time independent of n. All the answers a
re consistent with a single hyperfinite decomposition. In this talk\, I wi
ll describe a partition oracle that runs in time poly(d/\\epsilon). I will
also describe how this machinery strikes a fortune and helps in obtaining
a tester for all planar properties which runs in time exp(d/epsilon^2). T
his is easily obtained by a better error analysis on the seminal result of
Newman and Sohler [SICOMP 2013].\nBased on Joint works with Sabyasachi Ba
su\, C. Seshadhri and Andrew Stolman.\n
URL:https://www.tcs.tifr.res.in/web/events/1245
DTSTART;TZID=Asia/Kolkata:20221018T160000
DTEND;TZID=Asia/Kolkata:20221018T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1246
DTSTAMP:20230914T125956Z
SUMMARY:Total-payoff games on graphs with windows
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nWe look at a two-player ga
me played on a weighted directed graph. The game begins by placing a pawn
on a vertex in the graph. The players then take turns moving the pawn to a
n adjacent vertex in the graph\, yielding an infinite walk in the graph. E
ach time an edge is traversed\, Player 1 receives a payoff equal to the we
ight of the edge. An objective that has been traditionally studied for suc
h games is the total-payoff objective.\nIn this objective\, Player 1 wins
the game if the total payoff that they receive in the long run is non-nega
tive\, whereas Player 2 wins if the total-payoff is negative.\nA drawback
of this objective is that Player 1 may win the game\, and yet there may ex
ist arbitrarily long stretches in the play in which the total-payoff is ne
gative. This leads us to consider a stronger\nobjective: window total-payo
ff. In this objective\, Player 1 wins if from every point in the play\, th
ey can ensure that the total-payoff becomes non-negative in at most L step
s. This objective gives stronger guarantees on the payoff received by Play
er 1 while at the same time being easier to compute. Computing the winni
ng regions and winning strategies for the players for the traditional tota
l-payoff objective is only known to be in NP \\cap coNP\, whereas we see p
olynomial time algorithms to compute the same for the window total-payoff
objective.\nAll results presented in this talk are from the paper “Looki
ng at mean-payoff and total-payoff through windows” by Chatterjee\, Doye
n\, Randour\, and Raskin (2015).\n
URL:https://www.tcs.tifr.res.in/web/events/1246
DTSTART;TZID=Asia/Kolkata:20221021T140000
DTEND;TZID=Asia/Kolkata:20221021T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1247
DTSTAMP:20230914T125956Z
SUMMARY:Commitment over Unreliable Noisy Channels: When Awareness Meets Con
trol
DESCRIPTION:Speaker: Manideep Mamindlapally\n\nAbstract: \nWe study commitm
ents over unreliable compound noisy channels\, where parties may know the
compound channel state but lack control over that state. Commitment is one
of the common building blocks for many cryptographic protocols which real
ize multi-party computational functionalities in a secure manner. A noisy
channel\, among others\, is widely acknowledged as a promising resource fo
r realizing information-theoretically secure cryptographic primitives\, in
cluding commitment. However\, unreliable noisy channels with poorly or imp
recisely characterized statistical behaviour can severely degrade commitme
nt guarantees. Our focus is unreliability on account of a compound channel
state\, albeit under public awareness of that state. Building on prior wo
rk\, we seek to characterize the optimal commitment throughput or commitme
nt capacity of compound binary symmetric channels when parties may be stat
e-aware. State-awareness implies a passive and publicly known capability o
f precise channel knowledge (whether said party is honest or dishonest)\;
however\, state-awareness precludes any active and private channel state c
ontrol as in\, for instance\, the classic unfair noisy channels (UNCs). We
present new results on the commitment capacity under all possible configu
rations where individual parties may (or may not) be state-aware. An impor
tant takeaway of our work is the following: even a fairly weak capability
of state-awareness (albeit when asymmetric and only at the committer-side)
can degrade the commitment throughput to the same extent as under strongl
y capable parties that can privately control the compound state.\n
URL:https://www.tcs.tifr.res.in/web/events/1247
DTSTART;TZID=Asia/Kolkata:20221028T150000
DTEND;TZID=Asia/Kolkata:20221028T160000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1248
DTSTAMP:20230914T125956Z
SUMMARY:Logic is not logic
DESCRIPTION:Speaker: Jean-Yves Beziau (Federal University of Rio de Janeiro
\, Brazil.)\n\nAbstract: \nLogic is an important field of research with ma
ny aspects. In this talk we will discuss in particular the difference betw
een logic as reasoning and logic as a theory about reasoning. In the ligh
t of this distinction we will examine the unfolding of logic into logic a
nd metalogic\, the proliferation of logic systems\, the question of the r
elativity and the universality of logic and the place and interaction of l
ogic with regards to other sciences such as mathematics\, semiotics and c
omputer science. \n\nShort bio:\nJean-Yves Beziau is a Swiss Logician\, P
hilosopher and Mathematician\, PhD in mathematics and PhD in Philosophy. H
e has been living and working in different places: France\, Brazil\, Polan
d\, Corsica\, California (UCLA\, Stanford\, UCSD)\, Switzerland. He is cur
rently Professor at the University of Brazil in Rio de Janeiro\, former Di
rector of Graduate Studies in Philosophy and former President of the Brazi
lian Academy of Philosophy.\nHe is the creator of the World Logic Day\, ye
arly celebrated on January 14 (UNESCO international days)\, the World Logi
c Prizes Contest\, the founder and Editor-in-Chief of the journal Logica U
niversalis and South American Journal of Logic\, the book series Logic Ph
Ds\, Studies in Universal Logic and area logic editor of the Internet Enc
yclopedia of Philosophy. He has published about 200 research papers and
30 edited books and Special Issues of Journals. He has also organized many
events around the world\, having launched in particular four series of Wo
rld Congresses.\n
URL:https://www.tcs.tifr.res.in/web/events/1248
DTSTART;TZID=Asia/Kolkata:20221101T160000
DTEND;TZID=Asia/Kolkata:20221101T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1249
DTSTAMP:20230914T125956Z
SUMMARY:Best arm identification in the fixed confidence setting
DESCRIPTION:Speaker: Anirban Bhattacharjee\n\nAbstract: \nMulti-armed bandi
ts are sequential decision-making problems represented as a collection of
probability distributions that one can sample from at every time step. One
way to approach bandit problems is with the target of minimizing the expe
cted regret (penalty for not sampling from the distribution with the highe
st mean)\, given a fixed number of times that one can draw samples from th
is distribution (called the sampling budget). The second way of approachin
g these problems is with the target of minimizing the expected number of t
imes these distributions need to be sampled from (called the sampling comp
lexity)\, in order to declare the "best arm" with reasonable certainty\, g
iven the extent of certainty that is desired. We shall look at the "best a
rm" approach to multi-armed bandits when all the probability distributions
belong to a single parameter exponential family\, the lower bound on the
sampling complexity\, and how this lower bound may be asymptotically met.\
n
URL:https://www.tcs.tifr.res.in/web/events/1249
DTSTART;TZID=Asia/Kolkata:20221104T160000
DTEND;TZID=Asia/Kolkata:20221104T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1250
DTSTAMP:20230914T125956Z
SUMMARY:Maximally Recoverable Codes for Distributed Storage Systems
DESCRIPTION:Speaker: Lalitha Vadlamani (IIIT Hyderabad)\n\nAbstract: \nIn a
distributed storage system\, due to increase of storage capacity of a nod
e\, efficient repair of failed nodes is becoming increasingly important in
addition to ensuring a given level of reliability and low storage overhea
d. Codes with locality are a class of codes designed for storage systems w
hich have the characteristic that they trade off repair locality (number o
f nodes accessed to repair a failed node) for storage overhead.\n\nMaximal
ly recoverable codes are a class of codes which correct maximum possible n
umber of erasure patterns\, given the locality constraints of the code and
hence of interest. Two classes of maximally recoverable codes (MRC) based
on the topology of the local parities will be introduced (i) MRC with hie
rarchical locality (ii) MRCs with product topologies. For the case of MRC
with hierarchical locality\, we will present explicit constructions for al
l parameters and field size bounds. For the case of MRCs with product topo
logies\, we describe a certain regularity condition necessary for the eras
ure patterns to be recoverable. Also\, we establish a connection between t
he regularity condition and a complete matching in a suitably constructed
bipartite graph. This is joint work with D. Shivakrishna\, Aaditya M. Nair
\, V. Arvind Rameshwar and Birenjith Sasidharan.\nBio: Lalitha Vadlamani r
eceived the B.E. degree in electronics and communication engineering from
Osmania University\, Hyderabad\, in 2003\, and the M.E. and Ph.D. degrees
from the Indian Institute of Science (IISc)\, Bengaluru\, in 2005 and 2015
\, respectively. Since May 2015\, she has been working as an Assistant P
rofessor with IIIT Hyderabad\, where she is currently with the Signal Proc
essing and Communications Research Center. Her research interests include
coding for distributed storage and computing\, index coding\, polar codes\
, learning-based codes and coded blockchains. She was a recipient of Prof.
I. S. N. Murthy medal from IISc in 2005\, and the TCS Research Scholarshi
p for the year 2011. Her article won the runner up best paper award at NCC
2019.\n
URL:https://www.tcs.tifr.res.in/web/events/1250
DTSTART;TZID=Asia/Kolkata:20221111T113000
DTEND;TZID=Asia/Kolkata:20221111T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1251
DTSTAMP:20230914T125956Z
SUMMARY:On The Linear Arboricity Conjecture
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nA linear
forest is a forest in which every connected component is a line. The linea
r arboricity of an undirected graph is the minimum t such that the edge se
t can be partitioned into t subsets\, each of which forms a linear forest.
\n\nHarari introduced this quantity as one of the covering invariants of g
raphs. Harari\, Exoo and Akiyama (1981) made the following conjecture: the
linear arboricity of any d regular graph is ceil(d+1)/2 (the lower bound
is easy to prove\; the non-trivial part is the upper bound).\n\nAlon prove
d in 1988 that the linear arboricity of a d regular graph is at most d/2 +
O(d^{3/4+\\epsilon}) for any $\\epsilon > 0$. The proof uses Lovasz Local
Lemma and probabilistic method in very unexpected situations. We shall pr
esent this proof.\n\nLink to paper: https://link.springer.com/article/10.1
007/BF02783300\n
URL:https://www.tcs.tifr.res.in/web/events/1251
DTSTART;TZID=Asia/Kolkata:20221111T160000
DTEND;TZID=Asia/Kolkata:20221111T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1252
DTSTAMP:20230914T125956Z
SUMMARY:The Role of Adaptivity in Learning and Decision-Making
DESCRIPTION:Speaker: Arpit Agarwal (Columbia University)\n\nAbstract: \nIn
many machine learning applications the learner is faced with a *stochastic
environment* and it (sequentially) probes or influences the environment s
o as to optimize a given objective function. Examples of such applications
include recommendation systems\, web advertising\, viral marketing\, clin
ical trials\, search ranking etc. For instance\, in recommendation systems
\, the learner attempts to identify good recommendations by probing the st
ochastic preferences of users. Similarly\, in viral marketing\, the learne
r attempts to spread information through a social network using marketing
campaigns that influence (stochastic) subsets of the network.\n\nMost exis
ting learning algorithms for these applications operate in one of two sett
ings: (1) non-adaptive\, and (2) fully adaptive. In the non-adaptive setti
ng\, all the selections/probes are completely determined ahead of time. Ho
wever\, these a priori selections might be inefficient as some of them mig
ht be unnecessary in hindsight. In the fully adaptive setting\, the select
ion policy is updated after each observation from the environment. However
\, this fully adaptive setting might not be practical in many applications
due to delays in receiving observations from many parallel sources. In th
is talk we introduce a semi-adaptive setting that interpolates between the
se two extreme settings for a wide range of learning and decision-making p
roblems such as best arm identification in multi-armed bandits\, ranking f
rom pairwise comparisons\, dueling bandits\, and stochastic submodular max
imization. We show that semi-adaptive policies enjoy the power of fully ad
aptive policies while requiring very few updates to the selection/probing
rules. We also identify the trade-offs between rounds of adaptivity and pe
rformance.\n\nBio: Dr. Arpit Agarwal is a Postdoctoral Research Fellow at
the Data Science Institute at Columbia University. His research interests
primarily lie in machine learning\, with connections to algorithmic econom
ics and theoretical computer science. Before joining Columbia\, he receive
d a Ph.D. in Computer Science from the University of Pennsylvania and a ma
sters in Computer Science from the Indian Institute of Science.\n
URL:https://www.tcs.tifr.res.in/web/events/1252
DTSTART;TZID=Asia/Kolkata:20221114T160000
DTEND;TZID=Asia/Kolkata:20221114T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1253
DTSTAMP:20230914T125956Z
SUMMARY:How to Simulate it?
DESCRIPTION:Speaker: Manideep Mamindlapally (STCS\, TIFR\, Mumbai.)\n\nAbst
ract: \nThe notion of security for multi-party computations is not absolut
ely defined. Simulation is a technique that models an abstract "ideal worl
d" where a desired multi-party functionality is computed\, and is secure b
y definition. Then in the "real world" you would execute any given interac
tive protocol between agents\, who could be honest or cheating. You would
then argue that the outputs are similar to those of the secure "ideal worl
d"\, and in doing so prove that the protocol itself is secure. In this sem
inar\, I will talk about some such simulation proof techniques for some co
mmon multi-party computation problems - oblivious transfer and the zero-kn
owledge proof for the 3-colouring graph problem - based on examples from a
tutorial by Yehuda Lindell.\n
URL:https://www.tcs.tifr.res.in/web/events/1253
DTSTART;TZID=Asia/Kolkata:20221125T160000
DTEND;TZID=Asia/Kolkata:20221125T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1254
DTSTAMP:20230914T125956Z
SUMMARY:Median voter theorem and Why it Fails in Practice
DESCRIPTION:Speaker: Soumyajit Pyne\n\nAbstract: \nThe Median Voter Theorem
(MVT) is a cornerstone of modern political science theory. It states that
in a single peaked preference profile\, the top choice of the median vote
r is the Condorcet winner. So\, the dominant strategy for a candidate is t
o be the top choice of the median voter. However in practice\, often the m
edian voter's top choice is not the winner. To understand this we will go
through the paper "Polarization\, abstention\, and the median voter theore
m" - by Fu\, Sirianni\, Jones. Later I will show some generalization of th
e model used in that paper and analyze some computational aspects of that
model.\n
URL:https://www.tcs.tifr.res.in/web/events/1254
DTSTART;TZID=Asia/Kolkata:20221202T160000
DTEND;TZID=Asia/Kolkata:20221202T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1255
DTSTAMP:20230914T125956Z
SUMMARY:Bandits with Heavy Tails: Algorithms\, Analysis and Optimality
DESCRIPTION:Speaker: Shubhada Agrawal\n\nAbstract: \nMulti-armed bandit (MA
B) is a popular framework for sequential decision-making in an uncertain e
nvironment. In the classical setup of MAB\, the algorithm has access to a
fixed and finite set K of unknown\, independent probability distributions
or arms. At each time step\, having observed the outcomes of all the previ
ous actions\, the algorithm chooses one of the K arms and receives an inde
pendent sample drawn from the underlying distribution\, which may be consi
dered a reward. The algorithm's goal is either to maximize the accumulated
rewards or to identify the best arm in as few samples as possible for an
appropriate notion of best.\nVariants of these classical formulations have
been widely studied in the literature. Tight lower bounds and optimal alg
orithms have been developed under the assumption that the arm distribution
s are either Sub-Gaussian or come from a single parameter exponential fami
ly (SPEF)\, for example\, Gaussian distributions with a known variance or
Bernoulli distributions. However\, in practice\, these distributional assu
mptions may not hold. Developing lower bounds and optimal algorithms for t
he general distributions largely remained open\, mainly because of the nee
d for new tools for the analysis in this generality.\nIn this dissertation
\, we undertake a detailed study of the MAB problems allowing for all the
distributions with a known uniform bound on their (1+\\epsilon)^{th} momen
ts for some $\\epsilon > 0$. This class subsumes a large class of heavy-ta
iled distributions. We develop a framework with essential tools and concen
tration inequalities and use it to design optimal algorithms for three key
variants of the MAB problem\, including the classical frameworks of regre
t minimization and best-arm identification.\nA key component of designing
an optimal algorithm for MAB is constructing tight\, anytime valid confide
nce intervals (CIs) for mean. We develop new concentration inequalities to
this end\, which may be of independent interest.\nThe above results were
obtained in collaborations involving Sandeep Juneja\, Wouter M. Koolen and
Peter Glynn.\n\nZoom link: https://zoom.us/j/94294491152?pwd=MUZiNkk5Sy9
2Z3VEc3laZUNCTGdDdz09\n
URL:https://www.tcs.tifr.res.in/web/events/1255
DTSTART;TZID=Asia/Kolkata:20221209T113000
DTEND;TZID=Asia/Kolkata:20221209T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1256
DTSTAMP:20230914T125956Z
SUMMARY:Demystifying Approximate Reinforcement Learning Algorithms that use
epsilon-greedy Exploration
DESCRIPTION:Speaker: Aditya Gopalan (Indian Institute of Science (IISc)\nBe
ngaluru)\n\nAbstract: \nIn reinforcement learning\, value-function methods
such as Q-learning and SARSA(0) with $\\epsilon$-greedy exploration are a
mong the state of the art\, and their tabular (exact) forms converge to th
e optimal Q-function under reasonable conditions. However\, with function
approximation\, these methods are known to exhibit strange behaviors\, e.g
.\, policy oscillation and chattering\, convergence to different attractor
s (possibly even the worst policy) on different runs\, etc.\, apart from t
he well-known instability of iterates. Accordingly\, a theory to explain t
hese phenomena has been a long-standing open problem\, even for basic line
ar function approximation (Sutton\, 1999). Our work uses differential incl
usion theory to provide the first framework for resolving this problem. We
further illustrate via numerical examples how this framework helps comple
tely explain these algorithms' asymptotic behaviors. (Joint work with Guga
n Thoppe\, IISc)\n
URL:https://www.tcs.tifr.res.in/web/events/1256
DTSTART;TZID=Asia/Kolkata:20221209T143000
DTEND;TZID=Asia/Kolkata:20221209T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1257
DTSTAMP:20230914T125956Z
SUMMARY:Information Freshness for Monitoring and Control over Wireless Netw
orks
DESCRIPTION:Speaker: Vishrant Tripathi (Massachusetts Institute of Technolo
gy (MIT))\n\nAbstract: \nI will motivate information freshness as an essen
tial ingredient for performing monitoring and control tasks over wireless
networks. To this end\, I will introduce the notion of Age-of-Information
(AoI) which has become popular in networking literature over the past deca
de. I will show that networked monitoring and control can be viewed as equ
ivalent AoI optimization problems\, and discuss how to solve these problem
s in single and multi hop networks. Next\, I will discuss learning of AoI
cost functions (when the mapping between system performance and informatio
n freshness is unknown and time-varying). For the final part of the talk\,
I will discuss WiSwarm - a wireless system for communication between mult
iple UAVs and a central base station - that brings our ideas from theory t
o the real world.\n
URL:https://www.tcs.tifr.res.in/web/events/1257
DTSTART;TZID=Asia/Kolkata:20221219T160000
DTEND;TZID=Asia/Kolkata:20221219T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1258
DTSTAMP:20230914T125956Z
SUMMARY:Almost linear time algorithms for max-flow and more
DESCRIPTION:Speaker: Sushant Sachdeva (University of Toronto)\n\nAbstract:
\nWe give the first almost-linear time algorithm for computing exact maxim
um flows and minimum-cost flows on directed graphs. By well known reductio
ns\, this implies almost-linear time algorithms for several problems inclu
ding bipartite matching\, optimal transport\, and undirected vertex connec
tivity.\n\nOur algorithm is designed using a new Interior Point Method (IP
M) that builds the flow as a sequence of almost-linear number of approxima
te undirected minimum-ratio cycles\, each of which is computed and process
ed very efficiently using a new dynamic data structure.\n\nOur framework e
xtends to give an almost-linear time algorithm for computing flows that mi
nimize general edge-separable convex functions to high accuracy. This give
s the first almost-linear time algorithm for several problems including en
tropy-regularized optimal transport\, matrix scaling\, p-norm flows\, and
Isotonic regression.\n\nJoint work with Li Chen\, Rasmus Kyng\, Yang Liu\,
Richard Peng\, and Maximilian Probst Gutenberg.\n\nBio: Sushant Sachdeva
is a faculty member at the CS dept. at the University of Toronto and a Vec
tor Institute affiliate. He is interested in algorithms\, and its connecti
ons to optimization\, machine learning\, and statistics. His recent resear
ch focus has been the design of fast algorithms for graph problems. https:
//www.cs.toronto.edu/~sachdeva/\n
URL:https://www.tcs.tifr.res.in/web/events/1258
DTSTART;TZID=Asia/Kolkata:20221222T143000
DTEND;TZID=Asia/Kolkata:20221222T153000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1259
DTSTAMP:20230914T125956Z
SUMMARY:The Kikuchi Matrix Method
DESCRIPTION:Speaker: Pravesh Kothari (Carnegie Mellon University)\n\nAbstra
ct: \nIn this talk\, I will present a new method that reduces understandin
g an appropriate notion of girth in hypergraphs to unraveling the spectrum
of a "Kikuchi" matrix associated with the hypergraph. I will discuss thre
e applications of this technique: 1. Finding a refutation algorithm for sm
oothed instances of constraint satisfaction problems (obtained by randomly
perturbing the literal patterns in a worst-case instance with a small pro
bability) that matches the best running-time vs constraint-density trade-o
ffs for the significantly special and easier case of random CSPs\, 2. Conf
irming Feige's 2008 Conjecture that postulated an extremal girth vs densit
y trade-off (a.k.a. Moore bounds) for k-uniform hypergraphs that generaliz
es the Alon-Hoory-Linial Moore bound for graphs\, 3. Proving a cubic lower
bound on the block length of 3 query locally decodable codes improving on
the prior best quadratic lower bound from the early 2000s. Based on joint
works with Omar Alrabiyah (Berkeley)\, Tim Hsieh (CMU)\, Peter Manohar (C
MU)\, Sidhanth Mohanty (Berkeley)\, and Venkat Guruswami (Berkeley).\n
URL:https://www.tcs.tifr.res.in/web/events/1259
DTSTART;TZID=Asia/Kolkata:20221223T113000
DTEND;TZID=Asia/Kolkata:20221223T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1260
DTSTAMP:20230914T125956Z
SUMMARY:Heterogeneous arrival streams in a two-queue network
DESCRIPTION:Speaker: Agniv Bandyopadhyay\n\nAbstract: \nIn a concert queuin
g game\, users decide when to arrive at a bottleneck queue of constant cap
acity that opens at a specific time. Every user wants to simultaneously mi
nimize her waiting time and get served as early as possible. In previous i
ncarnations\, this game has been studied over a single queue and over netw
orks with specific topologies such as tandem networks\, trellis networks\,
and general feed-forward networks having multiple layers\, with users arr
iving at one extreme layer and traveling to the other extreme. A unique no
n-cooperative Nash Equilibrium profile has been identified in the case of
the single queue and tandem networks with multiple layers. Whereas for tre
llis networks and general feed-forward networks\, the existence of non-coo
perative Nash Equilibrium profiles has been proved. It has also been prove
n that if there are multiple such equilibriums\, they will all have an equ
al social cost. In this project\, we attempt to study the situation where
multiple classes of players are traveling through different paths in a que
uing network. For simplicity\, we consider two groups of players who inten
d to travel on two distinct paths in a two-queue network. We also assume t
hat all players in a group have identical preferences. In all these instan
ces\, we can find a unique non-cooperative Nash Equilibrium profile as lon
g as the two groups have different preferences. Also\, in equilibrium\, th
e two groups will choose their order of arrival depending on how their pop
ulation sizes compare. We also observe that whenever the two groups share
one queue in their path\, if the queue's capacity is under an identified t
hreshold\, the two groups will arrive in that queue in contiguous disjoint
intervals. On the other hand\, if it is strictly above that threshold\, t
he two groups will arrive together in that queue over some contiguous inte
rval in the equilibrium profile.\n
URL:https://www.tcs.tifr.res.in/web/events/1260
DTSTART;TZID=Asia/Kolkata:20221223T160000
DTEND;TZID=Asia/Kolkata:20221223T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1261
DTSTAMP:20230914T125956Z
SUMMARY:Convex influences and a quantitative Gaussian correlation inequalit
y
DESCRIPTION:Speaker: Anindya De (University of Pennsylvania)\n\nAbstract: \
nThe Gaussian correlation inequality (GCI)\, proven by Royen in 2014\, st
ates that any two centrally symmetric convex sets (say K and L) in the Gau
ssian space are positively correlated. We will prove a new quantitative ve
rsion of the GCI which gives a lower bound on this correlation based on th
e "common influential directions" of K and L. This can be seen as a Gaussi
an space analogue of Talagrand's well known correlation inequality for mon
otone functions. To obtain this inequality\, we propose a new approach\, b
ased on analysis of Littlewood type polynomials\, which gives a recipe to
transfer qualitative correlation inequalities into quantitative correlatio
n inequalities. En route\, we also give a new notion of influences for con
vex symmetric sets over the Gaussian space which has many of the propertie
s of influences from Boolean functions over the discrete cube. Much remain
s to be explored\, in particular\, about this new notion of influences for
convex sets.\n\nBased on joint work with Shivam Nadimpalli and Rocco Serv
edio.\n
URL:https://www.tcs.tifr.res.in/web/events/1261
DTSTART;TZID=Asia/Kolkata:20221227T160000
DTEND;TZID=Asia/Kolkata:20221227T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1262
DTSTAMP:20230914T125957Z
SUMMARY:Communication With Adversary Identification in Byzantine Multiple A
ccess Channels
DESCRIPTION:Speaker: Neha Sangwan\n\nAbstract: \nWe introduce the problem o
f determining the identity of a byzantine user (internal adversary) in a c
ommunication system. We consider a two-user discrete memoryless multiple a
ccess channel where either user may deviate from the prescribed behaviour.
Owing to the noisy nature of the channel\, it may be overly restrictive t
o attempt to detect all deviations. In our formulation\, we only require d
etecting deviations which impede the decoding of the non-deviating user's
message. When neither user deviates\, correct decoding is required. When o
ne user deviates\, the decoder must either output a pair of messages of wh
ich the message of the non-deviating user is correct or identify the devia
ting user. The users and the receiver do not share any randomness. The res
ults include a characterization of the set of channels where communication
is feasible\, and an inner and outer bound on the capacity region.\n \nTh
is is based on a joint work with Mayank Bakshi\, Bikash Kumar Dey and Vino
d M. Prabhakaran\, and was presented at ISIT\, 2021.\n
URL:https://www.tcs.tifr.res.in/web/events/1262
DTSTART;TZID=Asia/Kolkata:20221230T170000
DTEND;TZID=Asia/Kolkata:20221230T180000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1263
DTSTAMP:20230914T125957Z
SUMMARY:Private Optimization and Statistical Physics: Low-Rank Matrix Appro
ximation
DESCRIPTION:Speaker: Nisheeth Vishnoi (Yale university)\n\nAbstract: \nIn t
his talk\, I will discuss the following connections between private optimi
zation and statistical physics in the context of the low-rank matrix appro
ximation problem:\n1) An efficient algorithm to privately compute a low-ra
nk approximation and how it leads to an efficient way to sample from Haris
h-Chandra-Itzykson-Zuber densities studied in physics and mathematics\, an
d\n2) An improved analysis of the "utility" of the "Gaussian Mechanism"
for private low-rank approximation using Dyson Brownian motion.\n
URL:https://www.tcs.tifr.res.in/web/events/1263
DTSTART;TZID=Asia/Kolkata:20230103T140000
DTEND;TZID=Asia/Kolkata:20230103T150000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1264
DTSTAMP:20230914T125957Z
SUMMARY:Classical Verification of Quantum Computations
DESCRIPTION:Speaker: Manideep Mamindlapally (TIFR\, Mumbai.)\n\nAbstract: \
nQuantum computers are in general believed to be more powerful than classi
cal computers\, but it is not clear if they are powerful enough to solve p
roblems that a classical computer can’t even verify. Are there any tasks
which a malicious quantum agent can solve and claim an answer that go unc
hecked by classical parties? There were several attempts to find the answe
r to this question. In this talk\, I will discuss a recent development whi
ch was the discovery of a protocol for a classical computer to efficiently
verify the results of any efficient quantum computation by Urmila Mahadev
.\n
URL:https://www.tcs.tifr.res.in/web/events/1264
DTSTART;TZID=Asia/Kolkata:20230106T160000
DTEND;TZID=Asia/Kolkata:20230106T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1265
DTSTAMP:20230914T125957Z
SUMMARY:Scaling limits of stochastic optimization algorithms over large gra
phs
DESCRIPTION:Speaker: Raghav Somani (University of Washington)\n\nAbstract:
\nWasserstein gradient flows often arise from mean-field interactions amon
g exchangeable particles. In many interesting applications however\, the "
particles" are edge weights in a graph whose vertex labels are exchangeabl
e but not the edges themselves. Motivated by such graph optimization probl
ems we investigate the question of optimization of functions over this dif
ferent class of symmetries. Popular applications include training of large
computational graphs like (Deep) Neural Networks. This body of work shows
that discrete stochastic optimization algorithms over finite graphs have
a well-defined analytical scaling limit as the size of the network grows t
o infinity. The limiting space is that of graphons\, a notion introduced b
y Lovász and Szegedy to describe limits of dense graph sequences. The lim
iting curves are given by a novel notion of McKean-Vlasov equations on gra
phons and a corresponding notion of propagation of chaos holds. In the asy
mptotically zero-noise case\, the limit is a gradient flow on the metric s
pace of graphons. This is an attempt to generalize the Wasserstein calculu
s to higher-order exchangeable structures.\n
URL:https://www.tcs.tifr.res.in/web/events/1265
DTSTART;TZID=Asia/Kolkata:20230117T160000
DTEND;TZID=Asia/Kolkata:20230117T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1266
DTSTAMP:20230914T125957Z
SUMMARY:A survey on finding Kings in tournaments
DESCRIPTION:Speaker: Arghya Chakraborty\n\nAbstract: \nIn any tournament be
tween several participants\, transitivity is sometimes not satisfied (i.e.
A may defeat B\, B defeats C and C in turn defeats A). In any case\, we s
hall have to define a winner. We shall first define the notion of a 'king'
in a tournament graph and then show that one can find a king by organizin
g O(n^{3/2}) matches when there are n participants. Also\, we shall show t
hat at least \\Omega(n^{4/3}) matches need to be organized in the worst ca
se.\n \nThe results that I will go over are mostly from https://epubs.siam
.org/doi/abs/10.1137/S0097539702410053.\n
URL:https://www.tcs.tifr.res.in/web/events/1266
DTSTART;TZID=Asia/Kolkata:20230120T160000
DTEND;TZID=Asia/Kolkata:20230120T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1267
DTSTAMP:20230914T125957Z
SUMMARY:A constant lower bound for the union-closed sets conjecture
DESCRIPTION:Speaker: Dr. Justin Gilmer (Google Brain)\n\nAbstract: \nI will
introduce Frankl's. conjecture\, discuss prior approaches and why it is a
bit notorious. Then will discuss an information-theoretic approach that e
stablishes a constant lower bound for the maximum element frequency in uni
on-closed families of sets.\n
URL:https://www.tcs.tifr.res.in/web/events/1267
DTSTART;TZID=Asia/Kolkata:20230124T093000
DTEND;TZID=Asia/Kolkata:20230124T103000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1268
DTSTAMP:20230914T125957Z
SUMMARY:Estimating the size of union of sets in streaming model
DESCRIPTION:Speaker: Sourav Chakraborty (ISI Kolkata)\n\nAbstract: \nWe pre
sent a very simple and efficient sampling-based algorithm for estimating t
he union of sets in the streaming setting. Suppose we have a collection of
sets S_1\, . . . \, S_M subsets of T\, arriving one by one in a stream\;
the sets are not given explicitly to us but rather defined implicitly via
the following oracles: for each set\, we can know the size of the set\, ge
t a uniform sample from the set\, and given a point check whether it belon
gs to the set. The goal is to estimate the size of the union of the sets S
_1\, . . . \, S_M.\nWe present a simple algorithm that estimates the size
of the union\, upto a (1 + \\epsilon) factor\, in space complexity and upd
ate time complexity O(log(M)^2/\\epsilon^2).\nOur algorithm provides the f
irst algorithm with polynomial dependence on the dimension for Klee’s me
asure problem in streaming setting and independent of the stream size\, th
ereby settling the open problem of Woodruff and Tirthpura (PODS-12).\nThis
talk will be based on works with Kuldeep Meel and Vinodchandran (PODS21\,
PODS22 and ESA22).\n
URL:https://www.tcs.tifr.res.in/web/events/1268
DTSTART;TZID=Asia/Kolkata:20230127T160000
DTEND;TZID=Asia/Kolkata:20230127T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1269
DTSTAMP:20230914T125957Z
SUMMARY:An Approximate Generalization of the Okamura-Seymour Theorem
DESCRIPTION:Speaker: Nikhil Kumar (Hasso-Plattner institute\, Potsdam\, Ger
many.)\n\nAbstract: \nWe consider the problem of multicommodity flows in p
lanar graphs. Okamura and Seymour showed that if all the demands are incid
ent on one face\, then the cut-condition is sufficient for routing demands
. We consider the following generalization of this setting and prove an ap
proximate max flow-min cut theorem: for every demand edge\, there exists a
face containing both its end points. We show that the cut-condition is su
fficient for routing Ω(1) -fraction of all the demands. To prove this\, w
e give a L1-embedding of the planar metric which approximately preserves d
istance between all pair of points on the same face.\n
URL:https://www.tcs.tifr.res.in/web/events/1269
DTSTART;TZID=Asia/Kolkata:20230131T093000
DTEND;TZID=Asia/Kolkata:20230131T103000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1270
DTSTAMP:20230914T125957Z
SUMMARY:The Moore Bound for Irregular Graphs
DESCRIPTION:Speaker: Shanthanu Suresh Rai\n\nAbstract: \nWhat is the large
st number of edges in a graph of order n and girth g? For d-regular graphs
\, essentially the best known answer is provided by the Moore bound. This
result can be extended to cover irregular graphs as well.\n\nIn the talk\,
we will look at what the Moore bound is and some problems related to Moor
e graphs. Then we will try to generalize the Moore bound for irregular gra
phs.\n\nPaper: https://link.springer.com/article/10.1007/s003730200002\n
URL:https://www.tcs.tifr.res.in/web/events/1270
DTSTART;TZID=Asia/Kolkata:20230203T160000
DTEND;TZID=Asia/Kolkata:20230203T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1271
DTSTAMP:20230914T125957Z
SUMMARY:A THEORETICALLY TRACTABLE CROSS VALIDATION FRAMEWORK FOR SIGNAL DEN
OISING
DESCRIPTION:Speaker: Sabyasachi Chatterjee (University of Ilinois at Urbana
Champagne)\n\nAbstract: \nWe formulate a general cross validation framewo
rk for signal denoising. The general framework is then applied to nonparam
etric regression methods such as Trend Filtering and Dyadic CART. The resu
lting cross validated versions are then shown to attain nearly the same ra
tes of convergence as are known for the optimally tuned analogues. There d
id not exist any previous theoretical analyses of cross validated versions
of Trend Filtering or Dyadic CART. Our general framework is inspired by t
he ideas in Chatterjee and Jafarov (2015) and is potentially applicable to
a wide range of estimation methods which use tuning parameters. BIO: Saby
asachi Chatterjee is an Assistant Professor (from 2017 onwards) in the Sta
tistics Department at University of Illinois at Urbana Champaign. Most of
his research has been in Statistical Signal Processing. His is also intere
sted in Probability and all theoretical aspects of Machine Learning. He ob
tained his Phd in 2014 at Yale University and then was a Kruskal Instructo
r at University of Chicago till 2017.\n
URL:https://www.tcs.tifr.res.in/web/events/1271
DTSTART;TZID=Asia/Kolkata:20230210T140000
DTEND;TZID=Asia/Kolkata:20230210T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1272
DTSTAMP:20230914T125957Z
SUMMARY:A visual proof of Fermat's two square theorem
DESCRIPTION:Speaker: Malhar Ajit Managoli\n\nAbstract: \nFermat's two squa
re theorem states that: \nAn odd prime p can be written as a sum of two sq
uares if and only if p = 1 (mod 4) \nFurthermore\, such a solution is uniq
ue. Many proofs have been given of this fact\, but most of them are very t
ricky. \nA relatively simple proof of the existence part appeared in a su
rvey of proofs of Fermat's theorem by Alexander Spivak\, where the key poi
nt is presented in a visual manner\, which I will present. The proof of un
iqueness is much simpler\, and I will present it\, time permitting.\n
URL:https://www.tcs.tifr.res.in/web/events/1272
DTSTART;TZID=Asia/Kolkata:20230210T160000
DTEND;TZID=Asia/Kolkata:20230210T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1273
DTSTAMP:20230914T125957Z
SUMMARY:A Chernoff-Stein Lemma for adversarial hypothesis testing
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nIn the composite hypothesi
s testing setting\, the detector receives n i.i.d. samples either from a d
istribution p∈P or from a distribution q∈Q. It then decides the corr
ect set from which the samples were drawn.\nBrandao\, Harrow\, Lee and Per
es [BHLP] considered an adaptive generalization of this problem where the
choice of p∈P and q∈Q can change in each sample in some way that depen
ds arbitrarily on the previous samples. Initially\, one might think that t
his might confuse the detector and worsen the optimal error exponent. But\
, [BHLP] showed that the exponent remains the same and a simple maximum li
kelihood ratio test achieves it.\nLink to the paper: https://arxiv.org/abs
/1308.6702\n
URL:https://www.tcs.tifr.res.in/web/events/1273
DTSTART;TZID=Asia/Kolkata:20230217T160000
DTEND;TZID=Asia/Kolkata:20230217T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1274
DTSTAMP:20230914T125957Z
SUMMARY:Computational variants of the general Lax conjecture
DESCRIPTION:Speaker: Rafael Oliviera (University of Waterloo)\n\nAbstract:
\nHyperbolicity cones are convex semialgebraic sets generalizing both poly
hedral and spectrahedral cones\, the latter forming the basic geometric se
ts from linear and semidefinite programming. Hyperbolic polynomials\, whic
h give rise to these hyperbolicity cones\, have recently found application
s in several areas of mathematics\, statistical physics\, computer science
\, and optimization. The general Lax conjecture is a fundamental question
in real algebraic geometry and optimization: do hyperbolicity cones form a
strict generalization of spectrahedral cones?\nIn this talk\, we will giv
e an introduction to hyperbolic polynomials and their cones\, and raise se
veral computational questions related to these objects\, which blend algeb
raic complexity\, real algebraic geometry\, proof complexity and optimizat
ion.\n
URL:https://www.tcs.tifr.res.in/web/events/1274
DTSTART;TZID=Asia/Kolkata:20230220T143000
DTEND;TZID=Asia/Kolkata:20230220T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1275
DTSTAMP:20230921T105045Z
SUMMARY:Communication is Everything. Everything is Communication
DESCRIPTION:Speaker: Raghuvansh Saxena (Microsoft Research)\n\nAbstract: \n
Communication complexity is the study of how two or more parties with priv
ate inputs compute a function that depends on all their inputs. The scarce
resource is communication\, or the number of bits exchanged between the p
arties. What is amazing about this field is that it has applications not o
nly in areas where there is actual communication between parties\, such as
auction design and distributed computing\, but also in areas which superf
icially may seem completely unrelated to communication\, such as graph str
eaming and data structures. This is because bounds on communication can of
ten be translated into bounds on other resources of interest\, such as mem
ory and the number of wires.\nIn this talk\, I will cover my work in devel
oping and applying new communication complexity tools to mechanism design\
, streaming algorithms\, error-resilient circuits\, and interactive coding
\, with a special focus on the latter. Specifically\, I shall cover two of
my recent results [EKS20a] and [EKSZ22]\, that develop new codes resilien
t to a larger fraction of noise than the previous state-of-the-art. In the
case of [EKSZ22]\, I will also explain why our result opens a whole new p
aradigm for error correcting codes that was previously unexplored. No prio
r background will be assumed.\n
URL:https://www.tcs.tifr.res.in/web/events/1275
DTSTART;TZID=Asia/Kolkata:20230221T090000
DTEND;TZID=Asia/Kolkata:20230221T100000
LOCATION:Online with A-201 screening
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1276
DTSTAMP:20230914T125957Z
SUMMARY:Fair and Efficient Allocations under Sub-additive Valuations
DESCRIPTION:Speaker: Yeshwant Chandrakant Pandit\n\nAbstract: \nFair divis
ion of a set of resources among several agents is a commonly occurring pro
blem in many real-world settings. In this paper\, we will study the proble
m of allocating a set of indivisible goods among agents with sub-additive
valuations in a fair and efficient manner. Envy-Freeness up to any good (E
FX) is the most compelling notion of fairness in the context of indivisibl
e goods. Although the existence of EFX is not known beyond the simple case
of two agents with sub-additive valuations \, some good approximations of
EFX are known to exist\, namely 1/2-EFX allocation and EFX allocations wi
th bounded charity.\n\nIn this talk\, we will look at a polynomial time al
gorithm that outputs an allocation that satisfies either of the two approx
imations of EFX as well as achieve an O(n) approximation to the Nash welfa
re\n
URL:https://www.tcs.tifr.res.in/web/events/1276
DTSTART;TZID=Asia/Kolkata:20230224T160000
DTEND;TZID=Asia/Kolkata:20230224T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1277
DTSTAMP:20230914T125957Z
SUMMARY:Determinantal complexity of the power sum polynomial
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nWe will introduce the
notion of determinantal complexity\, one of the main characters in the VP
vs VNP question\, which is the algebraic analogue of the P vs NP question.
We will focus on a specific polynomial - the power sum polynomial - and s
ee proof sketches for an upper bound and a lower bound on its determinanta
l complexity. The lower bound will be from a paper by Alper\, Bogart and V
elasco. Prerequisites for the talk are basic linear algebra (rank\, etc) a
nd basic calculus (partial derivatives\, chain rule\, etc). There will be
some usage of tools from algebraic geometry but we won't see their proofs.
\n
URL:https://www.tcs.tifr.res.in/web/events/1277
DTSTART;TZID=Asia/Kolkata:20230303T163000
DTEND;TZID=Asia/Kolkata:20230303T173000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1278
DTSTAMP:20230914T125957Z
SUMMARY:Determinantal complexity (part 2)
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nI will complete a coup
le of proofs from last week's student seminar on determinantal complexity.
I will recall the required background for the proofs. On the way\, we wil
l learn a cute linear-algebraic fact. The talk / the proofs should be acce
ssible even if you missed last week's seminar.\n
URL:https://www.tcs.tifr.res.in/web/events/1278
DTSTART;TZID=Asia/Kolkata:20230310T160000
DTEND;TZID=Asia/Kolkata:20230310T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1279
DTSTAMP:20230914T125957Z
SUMMARY:Errors and Correction in Cumulative Knowledge
DESCRIPTION:Speaker: Madhu Sudan (Harvard John A. Paulson School of Enginee
ring and Applied Sciences)\n\nAbstract: \nSocietal accumulation of knowled
ge is a complex\, and arguably error-prone\, process. The correctness of n
ew units of knowledge depends not only on the correctness of the new reaso
ning\, but also on the correctness of old units that the new one builds on
. Left unchecked errors could completely ruin the validity of most of this
knowledge - so there must some error-correcting going on. What are the er
ror-corrections processes and how effective are they? In this talk\, we pr
esent a simple probabilistic process that aims to model such accumulation
of knowledge and study the persistence (or lack thereof) of errors. % Our
model for the generation of new units of knowledge is based on the prefere
ntial attachment growth model\, to which we additionally allow for injecti
on of errors. Furthermore\, the process includes checks aimed at catching
these errors. We investigate when effects of errors persist forever in the
system (with positive probability) and when they get rooted out completel
y by the checking process. The two basic parameters associated with the ch
ecking process are the {\\em probability} of conducting a check and the {\
\em depth} of the check. We show that errors are rooted out if checks are
sufficiently frequent and sufficiently deep. In contrast\, shallow or infr
equent checks are insufficient to root out errors. Based on the paper: "Is
This Correct? Let's Check!" with Omri Ben-Eliezer\, Dan Mikulincer and El
chanan Mossel (all at MIT).\n
URL:https://www.tcs.tifr.res.in/web/events/1279
DTSTART;TZID=Asia/Kolkata:20230314T160000
DTEND;TZID=Asia/Kolkata:20230314T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1280
DTSTAMP:20230914T125957Z
SUMMARY:The Implicit Bias of Gradient Descent on Separable Data
DESCRIPTION:Speaker: Santanu Das\n\nAbstract: \nWe examine gradient descent
on unregularised logistic regression problems\, with homogeneous linear p
redictors on linearly separable datasets. We show the predictor converges
to the direction of the max-margin (hard margin SVM) solution. The result
also generalizes to other monotone decreasing loss functions with an infim
um at infinity\, to multi-class problems\, and to training a weight layer
in a deep network in a certain restricted setting. Furthermore\, we show t
his convergence is very slow\, and only logarithmic in the convergence of
the loss itself. This can help explain the benefit of continuing to optimi
se the logistic or cross-entropy loss even after the training error is zer
o and the training loss is extremely small\, and\, as we show\, even if th
e validation loss increases. Our methodology can also aid in understanding
implicit regularisation in more complex models and with other optimisatio
n methods.\n
URL:https://www.tcs.tifr.res.in/web/events/1280
DTSTART;TZID=Asia/Kolkata:20230317T160000
DTEND;TZID=Asia/Kolkata:20230317T173000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1281
DTSTAMP:20230914T125957Z
SUMMARY:Monte Carlo tree search with advice
DESCRIPTION:Speaker: Debraj Chakraborty (Université Libre de Bruxelles)\n\
nAbstract: \nWe show how to combine techniques from formal methods and lea
rning for online computation of a strategy that aims at optimizing the exp
ected long-term reward in large systems modelled as Markov decision proces
ses. This strategy is computed with a receding horizon and using Monte Car
lo tree search (MCTS). We augment the MCTS algorithm with the notion of ad
vice which guides the search in the relevant part of the tree using exact
methods. Such advice can be symbolically written as a logical formula and
computed on-the-fly using model-checking tools.\n\nWe show that the classi
cal theoretical guarantees of the Monte Carlo tree search are still mainta
ined after this augmentation. To lower the latency of MCTS algorithms with
advice\, we propose to replace advice coming from exact algorithms with a
n artificial neural network. For this purpose\, we implemented an expert i
mitation framework to train the neural network in order to replace expert
advice with neural advice. To demonstrate the practical interest of our te
chniques\, we implemented the frameworks on different systems modelled as
MDPs.\n
URL:https://www.tcs.tifr.res.in/web/events/1281
DTSTART;TZID=Asia/Kolkata:20230321T160000
DTEND;TZID=Asia/Kolkata:20230321T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1282
DTSTAMP:20230914T125957Z
SUMMARY:Efficiency of non-truthful auctions under auto-bidding
DESCRIPTION:Speaker: Agniv Bandyopadhyay\n\nAbstract: \nAuto-bidding is a f
ramework of ad auctions where every advertiser can tell their long-term go
als\, such as budget\, target return on spend (RoS)\, etc.\, to an auto-bi
dding agent interface. With this information\, the interface bids on behal
f of the advertiser to a central agency running some mechanism (an allocat
ion rule and payment rule) to allocate the available ad slot to one of the
bidders. A natural question is: what choice of mechanism provides the mos
t efficient allocation in this auto-bidding world? Aggarwal et al. (2019)
proved that a second price auction incurs a price of anarchy (PoA) of 2\,
even with two bidders (i.e.\, social welfare at the equilibrium bidding po
licy is at least 1/2 times the optimal social welfare achievable). Later M
ehta (2022) proved that upon allowing randomized mechanisms\, one could ha
ve a truthful mechanism with a PoA of approx. 1.896 when restricted to ins
tances with only two bidders. Moreover\, they proved an impossibility resu
lt that any randomized truthful mechanism has a PoA >=2 as the # of bidder
s increases to infinity. So\, one natural question can be\, can one have a
lower PoA by allowing a non-truthful mechanism?\n\nLiaw et al. (2022) add
ressed this question and proved that\, even for non-truthful mechanisms wi
th some assumptions\, PoA >=2 as the # of bidders increases to infinity. T
hey showed that every deterministic mechanism (truthful and non-truthful)
satisfying some assumptions has a PoA >= 2\, even with two bidders\, and t
he bound is tight for first-price auctions. They also constructed a non-tr
uthful randomized variant of the first-price auction with PoA=1.8 over ins
tances with two bidders. Moreover\, they proved that after modifying the s
ame mechanism's payment rule to be truthful\, PoA increases to 1.9\, provi
ng that non-truthfulness helps in those instances. As long as time permits
\, we will go through as many results as possible proven in this paper.\n
URL:https://www.tcs.tifr.res.in/web/events/1282
DTSTART;TZID=Asia/Kolkata:20230324T143000
DTEND;TZID=Asia/Kolkata:20230324T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1283
DTSTAMP:20230914T125957Z
SUMMARY:Introduction in Topological Vector Spaces and Banach Spaces
DESCRIPTION:Speaker: Sourav Roy\n\nAbstract: \nWe will start basic definit
ions and examples. We will try to cover Duality in Banach Spaces\, Compact
operators. If time permits we will see some convexity results known "Se
paration Theorem" and "Hahn- Banach Theorem".\n
URL:https://www.tcs.tifr.res.in/web/events/1283
DTSTART;TZID=Asia/Kolkata:20230331T160000
DTEND;TZID=Asia/Kolkata:20230331T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1284
DTSTAMP:20230914T125957Z
SUMMARY:On Robustness for Linear Recurrence Sequences
DESCRIPTION:Speaker: Mihir Vahanwala (MPI-SWS\, Saarbrücken)\n\nAbstract:
\nThe Skolem\, Positivity\, and Ultimate Positivity problems for Linear Re
currence Sequences (LRS) are number-theoretic problems whose decidability
has been open for decades. They are known to be at least as hard as Diopha
ntine approximation\, an open number-theoretic problem. LRS have scientifi
c applications ranging from theoretical biology to software verification a
nd formal languages. Given the inherent imprecision and need for safety ma
rgins in the real world\, we consider the problems for LRS with the follow
ing notion of robustness: does the sequence satisfy the required property
despite small perturbations in the given initialisation? Although some int
erpretations of this notion yield problems that are still Diophantine hard
\, others can be shown to be decidable\, in PSPACE even! In this talk\, we
discuss how the decision procedure strings together remarkable and profou
nd results from computational algebra\, number theory\, and logic. The tec
hniques we discuss are indeed a thematic feature of modern progress in thi
s area\, and we briefly sketch how they have been extended to handle more
sophisticated semi-algebraic reachability sets\, and to tackle the problem
of invariant synthesis.\nBio: Mihir Vahanwala is a doctoral researcher a
t MPI-SWS\, Saarbrücken. He works in the Foundations of Algorithmic Verif
ication group led by Joël Ouaknine.\n
URL:https://www.tcs.tifr.res.in/web/events/1284
DTSTART;TZID=Asia/Kolkata:20230405T160000
DTEND;TZID=Asia/Kolkata:20230405T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1285
DTSTAMP:20230914T125957Z
SUMMARY:Minimization of weighted automaton
DESCRIPTION:Speaker: Mohit Upmanyu (TIFR\, Mumbai)\n\nAbstract: \nWe will d
efine a weighted automaton over a semiring\, give examples and explain how
to find a minimal weighted automaton (for certain rings like fields\, int
egers etc.)\, If time permits we will then discuss how this helps solve
the problem of checking whether two weighted automata are equal or not.\n
URL:https://www.tcs.tifr.res.in/web/events/1285
DTSTART;TZID=Asia/Kolkata:20230406T180000
DTEND;TZID=Asia/Kolkata:20230406T190000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1286
DTSTAMP:20230914T125957Z
SUMMARY:Dual Mirror Descent for Online Allocation Problems
DESCRIPTION:Speaker: Agniv Bandyopadhyay\n\nAbstract: \nWe will look at the
situation where an agent decides what action to take sequentially over a
finite time horizon to maximize revenue subject to resource consumption co
nstraints. Every time the revenue generated will be some concave function
of the action picked. Also\, the revenue function and resource consumption
will be sampled i.i.d. from some fixed distribution every time\, and that
distribution will be unknown to the agent. Balseiro\, Hu\, and Mirrokni
(2020) introduced a general class of algorithms for this problem\, which
achieves sublinear regret. The algorithm maintains a dual multiplier at ev
ery time step and uses online mirror descent to update it. Depending on
the regularizer chosen\, the policy can be equivalent to dual sub-gradient
descent and dual-multiplicative weight update. Moreover\, this framework
can achieve sublinear regret in repeated second-price auctions where\, unl
ike the online allocation problem\, bidders are unaware of their revenue f
unctions and resource consumption (payment) before deciding their bid.\n\n
Our discussion will be based on the contents of the following paper:\n"Dua
l Mirror Descent for Online Allocation Problems"-Santiago Balseiro\, Haiha
o Lu\, Vahab Mirrokni\, In: Proceedings of the 37th International Conferen
ce on Machine Learning\, PMLR 119:613-628\, 2020\n
URL:https://www.tcs.tifr.res.in/web/events/1286
DTSTART;TZID=Asia/Kolkata:20230421T160000
DTEND;TZID=Asia/Kolkata:20230421T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1287
DTSTAMP:20230914T125958Z
SUMMARY:Bounded Negativity and Harbourne constants on Algebraic Surfaces
DESCRIPTION:Speaker: Aditya Subramaniam (TIFR)\n\nAbstract: \nStudy of Alge
braic Surfaces is a classical subject. To understand surfaces\, it is natu
ral to study the various properties of the curves on them. Even now\, ther
e are several interesting conjectures about the properties of curves on su
rfaces. One such conjecture is the Bounded Negativity Conjecture (BNC).\nL
et X be a nonsingular projective surface. Bounded Negativity Conjecture (B
NC) says that there is an integer b(X)\, depending only on X\, such that t
he self-intersection C^2 is at least b(X) for every reduced curve C on X.
This conjecture is false in positive characteristic\, but it is open in ch
aracteristic zero\, except in trivial cases. Harbourne constants were defi
ned in an attempt to tackle this problem. I will introduce some of these i
deas in this talk.\n
URL:https://www.tcs.tifr.res.in/web/events/1287
DTSTART;TZID=Asia/Kolkata:20230428T160000
DTEND;TZID=Asia/Kolkata:20230428T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1288
DTSTAMP:20230914T125958Z
SUMMARY:Foundations of Lattice-based Cryptography
DESCRIPTION:Speaker: Rajendra Kumar (Weizmann institute)\n\nAbstract: \nPub
lic key cryptography is essential for internet security\, and RSA and Diff
ie-Hellman are the most widely used public-key cryptosystems for internet
traffic. However\, recent progress in building quantum computers threatens
RSA and Diffie-Hellman's security\, as they are vulnerable to quantum adv
ersaries. To address this\, organizations like the National Institute of S
tandards and Technology (NIST) and the European Telecommunications Standar
ds Institute (ETSI) have started standardizing and deploying cryptosystems
that are secure against quantum attacks. Recently\, NIST has chosen Kyb
er and Dilithium\, lattice-based candidates\, as primary algorithms for se
curity against quantum adversaries. The security of these cryptosystems cr
ucially relies on the assumption that the best-known algorithms for the la
ttice problems cannot be significantly improved.\nIn this talk\, I will di
scuss the connections between the security of lattice-based cryptosystems
and the hardness of lattice problems. I will talk about classical and quan
tum algorithms for lattice problems. I will also discuss the works on the
fine-grained security of lattice-based Crypto.\n
URL:https://www.tcs.tifr.res.in/web/events/1288
DTSTART;TZID=Asia/Kolkata:20230501T110000
DTEND;TZID=Asia/Kolkata:20230501T120000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1289
DTSTAMP:20230914T125958Z
SUMMARY:On Complexity measures of Boolean functions
DESCRIPTION:Speaker: Tulasi mohan Molli\n\nAbstract: \nBoolean functions ca
pture various problems and situations arising in computer science and othe
r areas. In this thesis\, we study boolean functions using three complexit
y measures namely Probabilistic degree\, decision tree depth and Fourier d
imension. The thesis and the talk are organized into three parts one for e
ach of these measures.\nIn the first part of the talk\, we will focus on t
he Probabilistic degree of OR over Reals. This is based on joint work with
Bhandari\, Harsha and Srinivasan. In this part\, we will look at the cons
truction of a Probabilistic Polynomial for OR over Reals\, which improves
on the previous best construction due to Toda-Ogiwara and Beigel\, Tarui\,
Reingold and Speilman. We will also look at a lower bound on the Probab
ilistic degree of OR which matches our upper bound construction in a restr
icted setting.\n\nIn the second part\, we will focus on decision trees of
Boolean functions after random restrictions.\nThis is based on joint work
with Harsha and Shankar. In this part\, we will look at the notion of the
criticality of a Boolean function which captures several properties of Boo
lean functions tightly. In our work\, we prove a switching lemma-like stat
ement for Boolean Formulas which settles a conjecture of Rossman.\nIn the
third part\, we will look at a bunch of complexity measures which arise ou
t of the Fourier representation of Boolean functions and study the relatio
nship between them. This is based on joint work with Chakraborty\, Mande
\, Mittal\, Paraashar and Sanyal. In this part\, we will focus on a couple
of upper bounds on the Fourier dimension in terms of Fourier sparsity\, w
eight\, Fourier max-entropy and Fourier max-rank entropy. We will also e
xhibit functions which match these bounds.\n
URL:https://www.tcs.tifr.res.in/web/events/1289
DTSTART;TZID=Asia/Kolkata:20230502T160000
DTEND;TZID=Asia/Kolkata:20230502T180000
LOCATION:Hybrid Mode (A-201)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1290
DTSTAMP:20230914T125958Z
SUMMARY:Towards Next-Generation ML/AI: Robustness\, Optimization\, Privacy
DESCRIPTION:Speaker: Krishna Pillutla (Google Research\, U.S.A.)\n\nAbstrac
t: \nTwo trends have taken hold in machine learning and artificial intelli
gence: a move to massive\, general-purpose\, pre-trained models as well as
a move to small\, on-device models trained on distributed data. Both thes
e disparate settings face some common challenges: a need for (a) robustnes
s to deployment conditions that differ from training\, (b) faster optimiza
tion\, and (c) protection of data privacy.\nAs a result of the former tren
d\, large language models have displayed emergent capabilities they have n
ot been trained for. Recent models such as ChatGPT have attained the abili
ty to generate remarkably human-like long-form text. I will describe Mauve
\, a measure to quantify this ability by measuring the gap between the dis
tribution of generated text and that of human-written text. I will highlig
ht its good empirical performance and present some statistical estimation
results.\nThe move to massively distributed on-device federated learning o
f models opens up new challenges due to the natural diversity of the under
lying user data and the need to protect its privacy. I will discuss how to
reframe the learning problem to make the model robust to natural distribu
tion shifts arising from deployment on diverse users who do not conform to
the population trends in a manner that admits a distributed optimization
algorithm with end-to-end differential privacy.\nTo conclude\, I will disc
uss my ongoing efforts and future plans to work toward the next generation
of ML/AI techniques by combining the best of both worlds. I will discuss
applications ranging from differentially private language models and text
generation to decentralized learning.\n\nBio: Krishna Pillutla is a visiti
ng researcher (postdoc) at Google Research\, USA in the Federated Learning
team. He obtained his Ph.D. at the University of Washington where he was
advised by Zaid Harchaoui and Sham Kakade. Before that\, he received his M
.S. from Carnegie Mellon University and B.Tech. from IIT Bombay where he w
as advised by Nina Balcan and J. Saketha Nath respectively. Krishna's rese
arch has been recognized by a NeurIPS outstanding paper award (2021) and a
JP Morgan Ph.D. fellowship (2019-20).\n
URL:https://www.tcs.tifr.res.in/web/events/1290
DTSTART;TZID=Asia/Kolkata:20230502T190000
DTEND;TZID=Asia/Kolkata:20230502T200000
LOCATION:Online
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1291
DTSTAMP:20230914T125958Z
SUMMARY:Towards a Principled Approach to Large Scale Data Analysis
DESCRIPTION:Speaker: Parthe Pandit (University of California\, San Diego)\n
\nAbstract: \nThe contemporary practice of data analysis is driven in larg
e part by deep neural networks. These complex models have had a tremendous
impact on a broad range of applications in science and engineering. Train
ing deep neural networks however\, remains an art with practitioners relyi
ng on heuristics and trial-&-error procedures. To make Data Science reliab
le and widely accessible\, we need to develop a theoretical foundation to
characterize the behavior of these complex models\, and come up with simpl
er substitutes firmly grounded in engineering principles.\nRecently\, a cl
assical model -- Kernel Methods -- has emerged as a framework to understan
d the behavior of deep neural networks\, following the discovery of the Ne
ural Tangent Kernel. This leads to a natural question as to whether Kernel
Methods can provide a simpler substitute to Deep Neural Networks\, while
achieving the same prediction performance and scalability. I will present
2 works which show progress in this direction:\n1. Recursive Feature Machi
nes: a new class of adaptive kernel methods that learn task-specific featu
res (arxiv.org/2212.13881 [1])\,\n2. EigenPro3: a new iterative algorithm
that enables scalable training of large kernel models (arxiv.org/2302.0260
5 [2])\, accepted at ICML 2023.\n\nBio: Parthe Pandit is a Simons postdoc
toral fellow with the Halıcıoğlu Data Science Institute at UC San Diego
. He obtained his Ph.D. in Electrical and Computer Engineering\, and M.S.
in Statistics both from UCLA\, and a B.Tech. + M.Tech. dual degree in Elec
trical Engineering with a minor in Computer Science from IIT Bombay.\nHis
research spans Machine Learning and Signal Processing with a focus on the
design and statistical analysis of iterative procedures for estimation and
inference in high dimensions. He has been a recipient of the Jack K. Wolf
student paper award at ISIT 2019\, and a Distinguished PhD dissertation f
inalist at UCLA ECE in 2022. He was a research intern with Amazon Search a
nd Amazon AWS working on Large Language Models\; and with Citadel LLC\, wo
rking on optimal trade execution in financial markets. Apart from Machine
Learning and Signal Processing\, he has also published articles in Graph T
heory\, Coding Theory\, Network Economics\, and planning EV charging infra
structure.\n
URL:https://www.tcs.tifr.res.in/web/events/1291
DTSTART;TZID=Asia/Kolkata:20230504T090000
DTEND;TZID=Asia/Kolkata:20230504T100000
LOCATION:Screening in A-201 with Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1292
DTSTAMP:20230914T125958Z
SUMMARY:Simulation Is All You Need
DESCRIPTION:Speaker: Akshit Kumar (Columbia University\, U.S.A.)\n\nAbstrac
t: \nMotivated by online matching markets and network revenue management (
NRM) problems with many types (e.g.\, fulfillment optimization)\, we study
dynamic spatial matching (DSM) in which supply and demand live in d dimen
sional space and need to be matched with each other dynamically. If demand
and supply have the same spatial distribution\, greedy matching suffices\
, and achieves average match distance of the same order as the distance to
the nearest neighbor. If demand and supply have different spatial distrib
utions\, the matching constraint has bite and greedy matching fails. We in
troduce a unifying and practical algorithmic principle for NRM and DSM dub
bed SOAR: Simulate\, Optimize\, Assign\, Repeat\, which repeatedly simulat
es the future to enable good matching decisions. Simulating one sample pat
h at each stage already enables SOAR to produce near optimal regret for th
e majority of NRM models in the literature\, and for DSM with non-atomic d
emand and supply distributions. For particularly challenging NRM and DSM m
odels\, SOAR with multiple simulated sample paths at each stage achieves n
ear optimal regret. This is joint work with Omar Besbes\, Yilun Chen\, Yas
h Kanoria and Wenxin Zhang based on two papers linked here and here.\n
URL:https://www.tcs.tifr.res.in/web/events/1292
DTSTART;TZID=Asia/Kolkata:20230511T160000
DTEND;TZID=Asia/Kolkata:20230511T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1293
DTSTAMP:20230914T125958Z
SUMMARY:Why Robust Generalization in Deep Learning is Difficult: Perspectiv
e of Expressive Power
DESCRIPTION:Speaker: Santanu Das\n\nAbstract: \nIt is commonly known that m
odern neural networks become sensitive when they get exposed to adversaria
l examples. To get rid of this problem\, people start to train neural netw
orks by using adversarial training algorithms and adversarial examples as
training data. However\, although the robust training error can be equal t
o near zero by using some methods\, but the robust generalization error re
mains high for all those existing adversarial training algorithms. My talk
will be based on the paper "Why Robust Generalization in Deep Learning is
Difficult: Perspective of Expressive Power " authored by Binghui Li\, Jik
ai Jin2\,Han Zhong\, John E. Hopcroft\, Liwei Wang3. In this paper\, the a
uthors provide a theoretical understanding of this phenomenon from the per
spective of expressive power for deep neural networks.\n
URL:https://www.tcs.tifr.res.in/web/events/1293
DTSTART;TZID=Asia/Kolkata:20230512T143000
DTEND;TZID=Asia/Kolkata:20230512T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1294
DTSTAMP:20230914T125958Z
SUMMARY:Broadcast Channel Synthesis
DESCRIPTION:Speaker: Malhar Ajit Managoli\n\nAbstract: \nIn a coordination
or channel synthesis problem\, parties have to use communication and shar
ed randomness to produce correlated random outputs. The question is\, how
much communication and/or shared randomness is required to accurately gene
rate correlation.\nIn this project\, we study the problem of three parties
synthesizing a noisy broadcast channel. The first party observes a random
variable X\, and sends a common message to the other two\, who have to pr
oduce random variables Y and Z according to a prescribed distribution.\n
URL:https://www.tcs.tifr.res.in/web/events/1294
DTSTART;TZID=Asia/Kolkata:20230515T170000
DTEND;TZID=Asia/Kolkata:20230515T180000
LOCATION:A238
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1295
DTSTAMP:20230921T105046Z
SUMMARY:Some results at the Intersection of Game Theory and Logic
DESCRIPTION:Speaker: Ramit Das (The Institute of Mathematical Sciences (IMS
c)\, Chennai)\n\nAbstract: \nWe shall address the issues of modelling or f
ormalising game theoretic properties like Nash Equilibrium\, Finite Impr
ovement Property\, Weak Acyclicity of various game forms in various kinds
of logic. We shall investigate the _expressive powers_ offered by each log
ic\, the _model checking_ theorems and also a _completeness_ proof of a
decidable logic variant. We hope that this investigation would have an i
mpact on the formalisation of game theory and its allied areas like comput
ational social choice theory.\n
URL:https://www.tcs.tifr.res.in/web/events/1295
DTSTART;TZID=Asia/Kolkata:20230518T110000
DTEND;TZID=Asia/Kolkata:20230518T120000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1296
DTSTAMP:20230914T125958Z
SUMMARY:Functionality of a Random Graph is Polylogarithmic
DESCRIPTION:Speaker: Pavel Dvorak\n\nAbstract: \nThe functionality of a gra
ph defined by [Alecu et al.: Graph functionality\, JCTB2021] is a paramete
r that generalizes graph degeneracy.\nInformally\, the functionality of a
graph G is minimal k such that in any induced subgraph H of G there is a v
ertex v in H and a set S of k vertices that the adjacency of v and any ver
tex u in H is determined by adjacency of u and S.\nI'll show that random g
raph G = G(n\,p) has functionality asymptotically at least log n and at mo
st log^3 n. The upper bound is a fresh new result\, I'll be happy to discu
ss its correctness.\nThis is joint work with L. Folwarczný\, M. Opler\, P
. Pudlák\, R. Šámal\, T. Vu.\n
URL:https://www.tcs.tifr.res.in/web/events/1296
DTSTART;TZID=Asia/Kolkata:20230526T160000
DTEND;TZID=Asia/Kolkata:20230526T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1297
DTSTAMP:20230914T125958Z
SUMMARY:New results on reasoning using transformers
DESCRIPTION:Speaker: Piyush Srivastava\n\nAbstract: \nIn the ML reading gro
up this Friday\, we will look at a circuit model (or "architecture") calle
d "transformer" that ostensibly underlies recent popular applications such
as ChatGPT\, Bard and friends. We\, however\, will focus on works (most
ly empirical as of now) that attempt to understand its power in the more r
estricted setting of learning to "reason" in the setting of a restricted p
roof system. Two examples we will look at are\n1) Straight Line Programs
\n2) Proofs in the Isabelle proof system.\nReferences:\nVaswani\, Ashish\,
Noam Shazeer\, Niki Parmar\, Jakob Uszkoreit\, Llion Jones\, Aidan N Gome
z\, Łukasz Kaiser\, and Illia Polosukhin. "Attention Is All You Need." In
Advances in Neural Information Processing Systems\, Vol. 30. Curran Assoc
iates\, Inc.\, 2017. https://papers.nips.cc/paper/2017/hash/3f5ee243547dee
91fbd053c1c4a845aa-Abstract.html.\nZhang\, Yi\, Arturs Backurs\, Sébastie
n Bubeck\, Ronen Eldan\, Suriya Gunasekar\, and Tal Wagner. "Unveiling Tra
nsformers with LEGO: A Synthetic Reasoning Task." arXiv\, February 17\, 20
23. http://arxiv.org/abs/2206.04301.\nMikuła\, Maciej\, Szymon Antoniak\,
Szymon Tworkowski\, Albert Qiaochu Jiang\, Jin Peng Zhou\, Christian Szeg
edy\, Łukasz Kuciński\, Piotr Miłoś\, and Yuhuai Wu. "Magnushammer: A
Transformer-Based Approach to Premise Selection." arXiv\, March 8\, 2023.
https://doi.org/10.48550/arXiv.2303.04488.\n
URL:https://www.tcs.tifr.res.in/web/events/1297
DTSTART;TZID=Asia/Kolkata:20230602T143000
DTEND;TZID=Asia/Kolkata:20230602T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1298
DTSTAMP:20230914T125958Z
SUMMARY:The AKS algorithm
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nAKS (Agra
wal-Kayal-Saxena) algorithm is the first known algorithm for primality tes
ting that is both provably efficient (not dependent on generalized Riemann
hypothesis) and deterministic. I shall present the analysis of this algor
ithm.\n\nPaper: https://annals.math.princeton.edu/2004/160-2/p12\n
URL:https://www.tcs.tifr.res.in/web/events/1298
DTSTART;TZID=Asia/Kolkata:20230602T160000
DTEND;TZID=Asia/Kolkata:20230602T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1299
DTSTAMP:20230914T125958Z
SUMMARY:Lower bounds for Polynomial Calculus with extension variables over
finite fields
DESCRIPTION:Speaker: Sasank Mouli (University of California\, San Diego)\n\
nAbstract: \nPropositional proof complexity aims to prove lower bounds aga
inst tautological Boolean formulae for increasingly strong proof systems\,
with the ultimate goal of separating NP and coNP. A major open problem in
this area is lower bounds for AC0[p]-Frege proofs. Since lower bounds for
the corresponding circuit class AC0[p] were proved by Razborov and Smolen
sky through algebraic means\, algebraic proof systems such as Nullstellens
atz (Beame et. al.) and Polynomial Calculus (Clegg et. al.) were introduce
d with the intention of better understanding AC0[p]-Frege. Lower bounds fo
r the former systems have been obtained but it has not led to much progres
s for the latter. In this talk we will show how to obtain lower bounds for
a weak version of Polynomial Calculus with extension variables\, a proof
system which with strong enough extension variables can simulate AC0[p]-Fr
ege. In particular we show that for every prime p and n > 0\, there exist
tautologies over O(n log n) variables of degree O(log n) such that any Pol
ynomial Calculus proof with o(n^2) extension variables\, each depending on
O(log n) original variables requires exponential size. This builds on a r
ecent work of Sokolov (STOC 2020) and is joint work with Russell Impagliaz
zo and Toniann Pitassi\, appearing in CCC 2023.\n
URL:https://www.tcs.tifr.res.in/web/events/1299
DTSTART;TZID=Asia/Kolkata:20230605T160000
DTEND;TZID=Asia/Kolkata:20230605T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1300
DTSTAMP:20230914T125958Z
SUMMARY:Inference\, Compression\, and Communication under Limited Data Acce
ss
DESCRIPTION:Speaker: Sahasranand Kodinthirapully Ramanadhan (Institut Polyt
echnique de Paris)\n\nAbstract: \nFirst\, we consider a two-party distribu
ted hypothesis testing problem for correlated Gaussian random variables. F
or a d-dimensional random vector X and a scalar Y\, where X and Y are join
tly Gaussian with an unknown correlation vector $\\rho$\, parties $\\mathc
al{P}_1$ and $\\mathcal{P}_2$ observe independent copies of X and Y\, resp
ectively. The parties seek to test if their observations are correlated or
not\, namely they seek to test if $\\|\\rho\\|_2$ exceeds $\\tau$ or is i
t 0. To that end\, they communicate interactively and declare the test out
put. We show that roughly order $d/\\tau^2$ bits of communication are suff
icient and necessary for resolving the distributed correlation testing pro
blem above. Furthermore\, we establish a lower bound of roughly $d^2/\\tau
^2$ bits for the communication needed for distributed estimation of $\\rho
$\, implying that distributed correlation testing requires less communicat
ion than distributed estimation. Towards the end\, we shall discuss briefl
y a streaming signal compression problem where access to samples is limite
d\, and another recent work wherein we provide partial resolution to a 27-
year-old conjecture regarding the capacity of queue channels by considerin
g a limited form of feedback.\nBio: Sahasranand is currently a postdoctora
l fellow at Telecom Paris. He has completed his PhD from ECE\, IISc in 202
2. His research interests include statistical inference\, information theo
ry\, and signal processing.\n
URL:https://www.tcs.tifr.res.in/web/events/1300
DTSTART;TZID=Asia/Kolkata:20230606T163000
DTEND;TZID=Asia/Kolkata:20230606T173000
LOCATION:Online
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1301
DTSTAMP:20230914T125958Z
SUMMARY:Stochastic Approximations of Sampling Algorithms
DESCRIPTION:Speaker: Dheeraj Nagaraj (Google AI\, Bangalore)\n\nAbstract: \
nWe consider stochastic approximations of sampling algorithms like Langevi
n Monte Carlo (pathwise approximation via random batches) and Stein Variat
ional Gradient Descent (approximation in the space of distributions). Thes
e algorithms are heavily deployed in Bayesian inference\, and the physical
sciences.\nWe first consider pathwise approximation in Stochastic Gradien
t Langevin Dynamics (SGLD)\, we show that the noise induced by the random
batches is approximately Gaussian (due to the Central Limit Theorem) while
the Brownian motion driving the algorithm is exactly Gaussian. We utili
ze this structure to provide improved guarantees for sampling algorithms u
nder significantly weaker assumptions. We then propose covariance correcti
on\, which rescales the brownian motion to approximately remove the random
batch error. We show that covariance corrected algorithms enjoy even bett
er convergence.\nWe then consider stochastic approximation in the space of
probability distributions to obtain a new particle discretization of Stei
n Variational Gradient Descent (SVGD)\, an interacting particle based samp
ling algorithm. We introduce and analyze Virtual Particle SVGD (VP-SVGD)\,
which enjoys provably rapid convergence to the target. Our rates provide
a double exponential improvement over the prior state of the art convergen
ce results for SVGD under mild conditions\, giving us the first provably f
ast variant of SVGD.\nBased on joint work with Aniket Das (Google) and Ana
nt Raj (INRIA and UIUC)\n
URL:https://www.tcs.tifr.res.in/web/events/1301
DTSTART;TZID=Asia/Kolkata:20230608T110000
DTEND;TZID=Asia/Kolkata:20230608T120000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1302
DTSTAMP:20230914T125958Z
SUMMARY:Randomness Requirements for Three-Secret Sharing
DESCRIPTION:Speaker: Hari Krishnan P A\n\nAbstract: \nConsider a secret sha
ring problem with three secrets related to each other\, i.e.\, we allow on
ly certain combinations of the three secrets. A dealer produces three shar
es such that every pair of share reveals a certain secret and nothing abou
t the other two secrets other than what can be inferred from the revealed
secret. We bound the randomness complexity of each possible set of permiss
ible combinations and give randomness-optimal schemes for the same for bin
ary secrets.\nThis is a joint work with Aayush Rajesh\, Varun Narayanan\,
Manoj Prabhakaran and Vinod Prabhakaran.\n
URL:https://www.tcs.tifr.res.in/web/events/1302
DTSTART;TZID=Asia/Kolkata:20230609T140000
DTEND;TZID=Asia/Kolkata:20230609T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1303
DTSTAMP:20230914T125958Z
SUMMARY:Differentiable Algorithms for Representation\, Processing and Rende
ring of Shapes
DESCRIPTION:Speaker: Aalok Gangopadhyay (IIT Gandhinagar)\n\nAbstract: \nOn
e of the primary objectives of visual computing has been the development o
f representations and algorithms that enable computer systems to acquire\,
process\, and render shapes efficiently. While numerous representations a
nd algorithms have been proposed\, many involve non-differentiable compone
nts\, making them incompatible with gradient-based optimization methods. T
he first part of this talk centers around developing differentiable repres
entations and rendering techniques for shapes. We present our work on desi
gning differentiable parameterized families of homeomorphisms and diffeomo
rphisms\, possibly with additional symmetry constraints\, which are used t
o deform a template shape having desired topological properties. This enab
les us to address shape search problems using gradient-based optimization
methods. We showcase our results for various applications\, including sear
ching for curve embeddings whose perspective projections resemble a target
image\, untangling knots\, parametrization of prototiles belonging to spe
cific isohedral tiling classes\, and density estimation on identification
spaces. In the subsequent part of this talk\, we focus on learning-based f
rameworks\, which generate differentiable algorithms for shape-processing
problems. Specifically\, we will elaborate on our work on surface denoisin
g\, surface correspondence\, temporal surface blending\, and curve extract
ion problems. The talk concludes with a discussion on future directions an
d potential areas of exploration for advancing the field.\n
URL:https://www.tcs.tifr.res.in/web/events/1303
DTSTART;TZID=Asia/Kolkata:20230612T160000
DTEND;TZID=Asia/Kolkata:20230612T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1304
DTSTAMP:20230921T105046Z
SUMMARY:DANSE - Data-driven Non-linear State Estimation in Unsupervised Lea
rning
DESCRIPTION:Speaker: Saikat Chatterjee (KTH Royal Institute of Technology\,
Stockholm\, Sweden)\n\nAbstract: \nWe address the tasks of Bayesian state
estimation and forecasting for a model-free process in an unsupervised le
arning setup. In the seminar\, we discuss our new method called DANSE -- D
ata-driven Nonlinear State Estimation method. DANSE provides a closed-form
posterior of the state of the model-free process\, given linear measureme
nts of the state. In addition it provides a closed-form posterior for fore
casting. We show how data-driven recurrent neural networks (RNNs) are used
in the DANSE to provide closed-form prior of the state and posterior. The
training of DANSE\, mainly learning the parameters of RNN\, is executed i
n unsupervised learning approach. In the unsupervised learning\, we have a
ccess to a training dataset comprising of only a set of measurement data t
rajectories\, but we do not have any access to the state trajectories. The
refore\, DANSE does not have access to state information in training data
and can not use supervised learning. Using simulated linear and non-linear
process models (Lorenz attractor and Chen attractor)\, we evaluate the un
supervised learning-based DANSE. We show that the proposed DANSE\, without
knowledge of the process model and without supervised learning\, provides
a competitive performance against model-driven methods\, such as Kalman f
ilter (KF)\, extended KF (EKF) and unscented KF (UKF)\, and a recently pro
posed hybrid method called KalmanNet.\n
URL:https://www.tcs.tifr.res.in/web/events/1304
DTSTART;TZID=Asia/Kolkata:20230616T160000
DTEND;TZID=Asia/Kolkata:20230616T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1305
DTSTAMP:20230914T125958Z
SUMMARY:Communication with Byzantine Users
DESCRIPTION:Speaker: Neha Sangwan\n\nAbstract: \nDistributed systems are ub
iquitous in modern society. These systems consist of a collection of nodes
or computers connected through a network. One of the fundamental requirem
ents of any distributed system is fault tolerance – even if some nodes a
re faulty or hacked\, it should not affect the functioning of the system a
s a whole. In this talk\, we will focus on two aspects of fault toleranc
e:\n(1) Is it possible to design a mechanism which allows different nodes
to reach consensus\, for example\, on a common state of the system or data
\, even in the presence of faulty nodes?\n(2) Can multiple nodes send info
rmation reliably over a shared communication medium\, even when some nodes
do not follow the protocol?\nTo answer the first question\, we focus on r
ealizing a cryptographic primitive called the byzantine broadcast\, which
ensures consensus among honest nodes in a network with one sender and mult
iple receivers. We use stochastic resources (like correlated randomness or
a noisy channel) to realize this primitive. For the second question\, we
model the setting using a noisy channel with multiple senders and a single
receiver (also called a multiple access channel) where some of the sender
s may maliciously deviate from the protocol. We study communication in thi
s model while ensuring that malicious senders are not able to cause undete
cted decoding errors for the honest senders.\n
URL:https://www.tcs.tifr.res.in/web/events/1305
DTSTART;TZID=Asia/Kolkata:20230621T113000
DTEND;TZID=Asia/Kolkata:20230621T130000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1306
DTSTAMP:20230914T125958Z
SUMMARY:Game of arrivals at a two queue network with heterogeneous route ch
oice
DESCRIPTION:Speaker: Agniv Bandyopadhyay\n\nAbstract: \nConsiderable resear
ch has focused on modelling customer arrival time to queues as games where
customers are strategic and non-cooperative in selecting their arrivals a
nd their costs are a function of the queueing delays as well as service co
mpletion times. Customers are typically modelled as fluid or non-atomic pa
rticles and services are assumed to have a deterministic rate. Most applic
ations involving strategic customers who think about when to join a queuin
g system fit this setting including customers arriving at a service provid
er(s)\, traffic networks and so on. Most of the literature has focused on
a single queue. Some works consider tandem and more general queuing networ
ks where customers have homogeneous routes. This homogeneity leads to subs
tantial analytical simplification. However\, there are many practical sett
ings where more than one queue is involved and customers have different ro
utes. Motivated by this\, we consider a queuing network that opens at a sp
ecified time\, where customers are non-atomic and belong to different clas
ses. Each class has its own route\, and as is typical in the literature\,
the costs are a linear function of waiting and service completion time. We
restrict ourselves to a two class\, two queue network: this simplificatio
n is well motivated as the diversity in solution structure as a function o
f problem parameters is substantial even in this simple setting (e.g.\, a
specific routing structure involves eight different regions)\, suggesting
a combinatorial blow up as the number of queues\, routes and customer clas
ses increase. We identify the unique Nash equilibrium customer arrival pro
file when the customer linear cost preferences are different. When custome
r cost preferences match\, under certain parametric settings\, the equilib
rium arrival profiles may not be unique and may lie in a convex set. We fu
rther make a surprising observation that in some parametric settings\, cus
tomers in one class may arrive in disjoint intervals\, although the union
of the arrival times across the two classes is always an interval. Further
\, the two classes may arrive in contiguous intervals or in overlapping in
tervals\, and at varying rates within an interval\, depending upon the pro
blem parameters. In this talk\, we will cover the existing works on single
queue setting and tandem networks. Following that\, if time permits\, we
will go through the analysis of the two queue settings.\nThis talk will be
based on joint work with Sandeep Juneja.\n
URL:https://www.tcs.tifr.res.in/web/events/1306
DTSTART;TZID=Asia/Kolkata:20230623T160000
DTEND;TZID=Asia/Kolkata:20230623T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1307
DTSTAMP:20230914T125958Z
SUMMARY:Counting Markov equivalence directed cyclic graphs consistent with
background knowledge
DESCRIPTION:Speaker: Vidya Sagar Sharma\n\nAbstract: \nWe study the problem
of counting the number of directed acyclic graphs in a Markov equivalence
class (MEC) that are consistent with background knowledge specified in t
he form of the directions of some additional edges in the MEC. A polynom
ial-time algorithm for the special case of the problem\, when no backgroun
d knowledge constraints are specified\, was given by Wienöbst\, Bannach\,
and Liśkiewicz (AAAI 2021)\, who also showed that the general case is NP
-hard (in fact\, \\#P-hard). In this talk\, we show that the problem is
nevertheless tractable in an interesting class of instances\, by establish
ing that it is ``fixed-parameter tractable'' (FPT): we give an algorithm t
hat runs in time $O(k! k^2 n^4)$\, where $n$ is the number of nodes in the
MEC and $k$ is the maximum number of nodes in any maximal clique of the M
EC that participate in the specified background knowledge constraints. I
n particular\, our algorithm runs in polynomial time in the well-studied s
pecial case of MECs of bounded tree-width or bounded maximum clique size.
\nPaper link: https://arxiv.org/abs/2206.06744\n
URL:https://www.tcs.tifr.res.in/web/events/1307
DTSTART;TZID=Asia/Kolkata:20230626T170000
DTEND;TZID=Asia/Kolkata:20230626T180000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1308
DTSTAMP:20230914T125958Z
SUMMARY:Tree-like Resolution and Decision Trees
DESCRIPTION:Speaker: Farzan Byramji (IIT Kanpur\, UCSD)\n\nAbstract: \nIt i
s well-known that tree-like resolution refutations correspond to decision
trees solving the associated false clause search problem. We describe this
connection and some lower bounds using this. Then we discuss variants of
tree-like resolution (resolution with parities and AND-resolution) along w
ith the corresponding decision tree models\, and lifting in this setting.\
n
URL:https://www.tcs.tifr.res.in/web/events/1308
DTSTART;TZID=Asia/Kolkata:20230629T160000
DTEND;TZID=Asia/Kolkata:20230629T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1309
DTSTAMP:20230914T125958Z
SUMMARY:New results on reasoning using transformer\, Part 2
DESCRIPTION:Speaker: Piyush Srivastava\n\nAbstract: \nIn the ML reading gro
up this Friday\, we will look at a circuit model (or "architecture") calle
d "transformer" that ostensibly underlies recent popular applications such
as ChatGPT\, Bard and friends. We\, however\, will focus on works (mostl
y empirical as of now) that attempt to understand its power in the more re
stricted setting of learning to "reason" in the setting of a restricted pr
oof system. Two examples we will look at are\n\n1) Straight Line Programs
\n2) Proofs in the Isabelle proof system.\n\nReferences:\n Vaswani\, Ashis
h\, Noam Shazeer\, Niki Parmar\, Jakob Uszkoreit\, Llion Jones\, Aidan N G
omez\, Łukasz Kaiser\, and Illia Polosukhin. "Attention Is All You Need."
In Advances in Neural Information Processing Systems\, Vol. 30. Curran As
sociates\, Inc.\, 2017. https://papers.nips.cc/paper/2017/hash/3f5ee243547
dee91fbd053c1c4a845aa-Abstract.html.\n Zhang\, Yi\, Arturs Backurs\, Séba
stien Bubeck\, Ronen Eldan\, Suriya Gunasekar\, and Tal Wagner. "Unveiling
Transformers with LEGO: A Synthetic Reasoning Task." arXiv\, February 17\
, 2023. http://arxiv.org/abs/2206.04301.\nMikuła\, Maciej\, Szymon Antoni
ak\, Szymon Tworkowski\, Albert Qiaochu Jiang\, Jin Peng Zhou\, Christian
Szegedy\, Łukasz Kuciński\, Piotr Miłoś\, and Yuhuai Wu. "Magnushammer
: A Transformer-Based Approach to Premise Selection." arXiv\, March 8\, 20
23. https://doi.org/10.48550/arXiv.2303.04488.\n
URL:https://www.tcs.tifr.res.in/web/events/1309
DTSTART;TZID=Asia/Kolkata:20230630T143000
DTEND;TZID=Asia/Kolkata:20230630T153000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1310
DTSTAMP:20230914T125958Z
SUMMARY:Optimal social choice functions: A utilitarian view
DESCRIPTION:Speaker: Soumyajit Pyne\n\nAbstract: \nThe authors adopt a util
itarian perspective on social choice\, assuming that agents have (possibly
latent) utility functions over some space of alternatives. For many reaso
ns one might consider mechanisms\, or social choice functions\, that only
have access to the ordinal rankings of alternatives by the individual agen
ts rather than their utility functions. In this context\, one possible obj
ective for a social choice function is the maximization of (expected) soci
al welfare relative to the information contained in these rankings. In the
talk\, we will focus on the worst case model (distortion) of the paper.\n
The link of the paper: https://www.sciencedirect.com/science/article/pii/S
0004370215000892\n
URL:https://www.tcs.tifr.res.in/web/events/1310
DTSTART;TZID=Asia/Kolkata:20230707T160000
DTEND;TZID=Asia/Kolkata:20230707T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1311
DTSTAMP:20230914T125958Z
SUMMARY:The unreasonable effectiveness of mathematics in large scale deep l
earning
DESCRIPTION:Speaker: Greg Yang (Microsoft New England Research and Developm
ent Center\, USA)\n\nAbstract: \nRecently\, the theory of infinite-width n
eural networks led to the first technology\, muTransfer\, for tuning enorm
ous neural networks that are too expensive to train more than once. For ex
ample\, this allowed us to tune the 6.7 billion parameter version of GPT-3
using only 7% of its pretraining compute budget\, and with some asterisks
\, we get a performance comparable to the original GPT-3 model with twice
the parameter count. In this talk\, I will explain the core insight behind
this theory. In fact\, this is an instance of what I call the Optimal Sca
ling Thesis\, which connects infinite-size limits for general notions of "
size" to the optimal design of large models in practice. I'll end with sev
eral concrete key mathematical research questions whose resolutions will h
ave an incredible impact on the future of AI.\n
URL:https://www.tcs.tifr.res.in/web/events/1311
DTSTART;TZID=Asia/Kolkata:20230712T110000
DTEND;TZID=Asia/Kolkata:20230712T123000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1312
DTSTAMP:20230914T125958Z
SUMMARY:A Sanov-type theorem for unimodular marked random graphs and its ap
plications
DESCRIPTION:Speaker: Sarath Yasodharan (Brown University\, U.S.A.)\n\nAbstr
act: \nWe prove a Sanov-type large deviation principle for the component e
mpirical measures of certain sequences of unimodular random graphs (includ
ing Erdos-Renyi and random regular graphs) whose vertices are marked with
i.i.d. random variables. Specifically\, we show that the rate function can
be expressed in a fairly tractable form involving suitable relative entro
py functionals. As a corollary\, we establish a variational formula for th
e annealed pressure (or limiting log partition function) for various stati
stical physics models on sparse random graphs.\nJoint work with I-Hsun Che
n and Kavita Ramanan.\n
URL:https://www.tcs.tifr.res.in/web/events/1312
DTSTART;TZID=Asia/Kolkata:20230712T160000
DTEND;TZID=Asia/Kolkata:20230712T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1313
DTSTAMP:20230914T125959Z
SUMMARY:Strategic Decision-Making for Information Freshness
DESCRIPTION:Speaker: Kumar Saurav\n\nAbstract: \nClassical network scheduli
ng problems have primarily focused on optimizing metrics\, such as delay\,
which pertain to the service provided to individual packets in the networ
k. However\, in modern applications like tele-robotics and networked cars\
, the emphasis is on metrics that capture the freshness of information\, s
pecifically\, how up-to-date the information is at the receiver (monitor)
compared to the transmitter (source). Thus\, several metrics have been int
roduced to quantify information freshness\, the most widely used one being
the age of information (AoI). The AoI for a source at any given time is e
qual to the difference between the current time and the generation time of
the most recent packet (update) received at the monitor. For modern app
lications\, the scheduling objective is to minimize the AoI for the source
s in an online environment\, where only causal information is available at
any time.\nA critical feature of AoI scheduling problems is that not all
updates generated at a source need to be transmitted. Depending on the net
work model\, a scheduling algorithm (policy) must choose a subset of updat
es to transmit. This characteristic gives AoI scheduling problems a combin
atorial flavour\, making them fundamentally different and analytically cha
llenging compared to classical scheduling problems. In the PhD dissertatio
n\, we have addressed some major challenges in AoI scheduling for generic
network models\, taking into consideration energy consumption\, shared tra
nsmission link\, and the type of scheduler (centralized/decentralized). We
have analyzed the various trade-offs involved in decision-making and prop
osed novel causal algorithms (policies) that strategically handle these tr
ade-offs. Additionally\, using analytical techniques\, we have derived the
oretical performance guarantees\, ensuring the efficiency and robustness o
f the proposed solutions. In this seminar\, we will have a comprehensive o
verview of the above contributions\, with all the relevant background.\n
URL:https://www.tcs.tifr.res.in/web/events/1313
DTSTART;TZID=Asia/Kolkata:20230713T160000
DTEND;TZID=Asia/Kolkata:20230713T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1314
DTSTAMP:20230914T125959Z
SUMMARY:Majority-3SAT in Polynomial Time
DESCRIPTION:Speaker: Ratnakar Medepalli\n\nAbstract: \nMajority-SAT is the
problem of determining whether a boolean formula evaluates to true on more
than half of its assignments. Majority-SAT is known to be PP-hard. In thi
s talk\, we will see a polynomial time algorithm for Majority-3SAT\, which
is the problem of determining whether a 3-CNF formula evaluates to true o
n more than half of its assignments. The result is surprising because most
SAT-related problems remain hard when restricted to their corresponding 3
-CNF variant.\nThis is a result of Shyan Akmal and Ryan Williams from FOCS
2021.\n
URL:https://www.tcs.tifr.res.in/web/events/1314
DTSTART;TZID=Asia/Kolkata:20230714T140000
DTEND;TZID=Asia/Kolkata:20230714T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1315
DTSTAMP:20230914T125959Z
SUMMARY:Robust Mean Estimation of Gaussian Random Variables
DESCRIPTION:Speaker: Malhar Ajit Managoli\n\nAbstract: \nSuppose we are gi
ven a set of n i.i.d. samples from a Gaussian with known variance but unkn
own mean. We wish to estimate the mean. It is well known that the sample m
ean is an excellent estimator for the true mean.\nHowever\, if a small fra
ction of the samples are corrupted (i.e. chosen by an adversary instead of
being drawn from the Gaussian)\, the sample mean can fail spectacularly.\
nIn this talk\, I will present an estimator which is robust to such corrup
tions and also show a lower bound.\n
URL:https://www.tcs.tifr.res.in/web/events/1315
DTSTART;TZID=Asia/Kolkata:20230714T160000
DTEND;TZID=Asia/Kolkata:20230714T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1316
DTSTAMP:20230914T125959Z
SUMMARY:Strong bounds for three-term progressions
DESCRIPTION:Speaker: Raghu Meka (University of California\, Los Angeles)\n\
nAbstract: \nSuppose you have a set $S$ of integers from $\\{1\,2\,...\,N\
\}$ that contains at least $N / C$ elements. Does such a set contain three
equally spaced numbers (i.e.\, a 3-term arithmetic progression)? For exam
ple\, the set $S$ from $\\{1\,2\,...\, 40\\}$ comprised of the following
15 numbers $\\{1\,2\,4\,5\,10\,11\,13\,14\,28\,29\,31\,32\,37\,38\,40\\}$
avoids every 3-term arithmetic progression.\nWhat is the largest value of
$C$ such that for large enough $N$\, every such set $S$ necessarily conta
ins a 3-term arithmetic progression?\nIn 1953\, Roth showed this is the ca
se when $C$ is roughly (log log N). Behrend in 1946 showed that $C$ can be
at most $exp(\\sqrt(\\log N))$. Since then\, the problem has been a corne
rstone of the area of additive combinatorics. Following a series of remark
able results\, a celebrated paper from 2020 due to Bloom and Sisask improv
ed the lower bound on $C$ to $C = (\\log N)^{1+c}$ for some constant $c >
0$.\nThis talk will describe a new work showing that $C$ can be as big as
$exp((log N)^{0.08})$\, thus getting closer to Behrend's construction. Bas
ed on joint work with Zander Kelley (Univ of Illinois\, Urbana Champaign).
\n
URL:https://www.tcs.tifr.res.in/web/events/1316
DTSTART;TZID=Asia/Kolkata:20230719T160000
DTEND;TZID=Asia/Kolkata:20230719T170000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1317
DTSTAMP:20230914T125959Z
SUMMARY:Strong bounds for three-term progressions (Part II)
DESCRIPTION:Speaker: Raghu Meka (University of California\, Los Angeles)\n\
nAbstract: \nSuppose you have a set $S$ of integers from $\\{1\,2\,...\,N\
\}$ that contains at least $N / C$ elements. Does such a set contain three
equally spaced numbers (i.e.\, a 3-term arithmetic progression)? For exam
ple\, the set $S$ from $\\{1\,2\,...\, 40\\}$ comprised of the following
15 numbers $\\{1\,2\,4\,5\,10\,11\,13\,14\,28\,29\,31\,32\,37\,38\,40\\}$
avoids every 3-term arithmetic progression.\nWhat is the largest value of
$C$ such that for large enough $N$\, every such set $S$ necessarily conta
ins a 3-term arithmetic progression?\nIn 1953\, Roth showed this is the ca
se when $C$ is roughly (log log N). Behrend in 1946 showed that $C$ can be
at most $exp(\\sqrt(\\log N))$. Since then\, the problem has been a corne
rstone of the area of additive combinatorics. Following a series of remark
able results\, a celebrated paper from 2020 due to Bloom and Sisask improv
ed the lower bound on $C$ to $C = (\\log N)^{1+c}$ for some constant $c >
0$.\nThis talk will describe a new work showing that $C$ can be as big as
$exp((log N)^{0.08})$\, thus getting closer to Behrend's construction. Bas
ed on joint work with Zander Kelley (Univ of Illinois\, Urbana Champaign).
\n
URL:https://www.tcs.tifr.res.in/web/events/1317
DTSTART;TZID=Asia/Kolkata:20230721T110000
DTEND;TZID=Asia/Kolkata:20230721T120000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1318
DTSTAMP:20230914T125959Z
SUMMARY:A Formal Approach to Exchange Design and Regulation
DESCRIPTION:Speaker: Suneel Sarswat\n\nAbstract: \nExchanges such as that o
f foreign currency\, stocks\, and commodities are organized marketplaces w
here trades are conducted by matching buy and sell requests of traders. Se
veral instances have been reported where exchanges have been found violati
ng regulatory guidelines and the stated rules. In this work\, we propose a
robust approach to exchange design and regulation by presenting a compreh
ensive framework for formalizing and certifying double auctions\, which ar
e the key mechanisms employed by the exchanges to match the buy and sell r
equests of traders. Typically\, two types of double auctions are employed
in the exchanges: call auctions and continuous auctions. For call auctions
again\, there are two main alternatives: uniform price and dynamic pric
e. Our main contributions are as follows.\n-- Call auctions: We formalize
the various notions of call auctions and provide fully formalized algorith
ms for uniform and dynamic price auctions along with their correctness pro
ofs.\n-- Continuous auctions: We formulate the specifications for continuo
us double auctions by identifying three simple and necessary properties an
d formally proving that they are in fact sufficient\; they completely dete
rmine the input-output relationship. We then formally verify that a natura
l algorithm satisfies these properties.\n-- Uniqueness theorems: We establ
ish new uniqueness theorems for both call and continuous auctions that ena
ble us to build automated checkers that are guaranteed to detect errors in
the trade logs of exchanges if they generate transactions that violate th
e specifications. We extract verified programs of our formalized algorithm
s to build automated checkers.\n-- Tests on real data: We add preprocessin
g steps that enable us to tailor our general model to a specific exchange.
We then run our automated checkers on real data. Furthermore\, we report
the running times on various input sizes.\n-- Efficiency: We obtain tight
bounds on the time complexity of all these three matching problems in the
comparison model. Specifically\, we demonstrate that uniform price matchin
g can be achieved in linear time\, which is an improvement over the previo
us algorithm that takes O(n log n) time to match n requests. For dynamic p
rice matching\, we establish a lower bound of Ω(n log n) on the running t
ime\, thereby proving that the currently known best algorithm is time-effi
cient. Furthermore\, for continuous double auctions\, we show that a natur
al algorithm takes O(n log n) time\, while any algorithm requires at least
Ω(n log n) time.\n
URL:https://www.tcs.tifr.res.in/web/events/1318
DTSTART;TZID=Asia/Kolkata:20230721T140000
DTEND;TZID=Asia/Kolkata:20230721T150000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1319
DTSTAMP:20230914T125959Z
SUMMARY:Fast Algorithms for Regression Problems
DESCRIPTION:Speaker: Deeksha Adil (ETH Zurich\, Switzerland)\n\nAbstract: \
nIncreasing data sizes necessitate fast and efficient algorithms for analy
zing them. Regression is one such essential tool that is used widely in co
mputer science. In this talk\, I will focus on the "p-norm regression prob
lem"\, which is a generalization of the standard "linear regression proble
m"\, and captures several important problems including the maximum flow pr
oblem on graphs. Historically\, obtaining fast\, high-accuracy algorithms
for this problem has been challenging due to the lack of smoothness and st
rong convexity of the function\, however\, recent breakthroughs have been
able to get around these issues. I will present an overview of how these a
lgorithms work and discuss some generalizations of these techniques to oth
er regression problems.\n
URL:https://www.tcs.tifr.res.in/web/events/1319
DTSTART;TZID=Asia/Kolkata:20230724T160000
DTEND;TZID=Asia/Kolkata:20230724T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1320
DTSTAMP:20230914T125959Z
SUMMARY:Sampling Algorithms via random walks on polytopes
DESCRIPTION:Speaker: Sourav Roy\n\nAbstract: \nGiven a polytope K of R^n\,
the problem of sampling from the uniform distribution over K arises in va
rious contexts. Till now there are various types of random walks proposed
on K. In this talk we will talk about Dikin Walk\, Vaidya Walk\, John Walk
defined on K.\n
URL:https://www.tcs.tifr.res.in/web/events/1320
DTSTART;TZID=Asia/Kolkata:20230728T160000
DTEND;TZID=Asia/Kolkata:20230728T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1321
DTSTAMP:20230921T105046Z
SUMMARY:A small numerical study about violation of additivity of Holevo cap
acity for sending classical information through quantum channels
DESCRIPTION:Speaker: Sai Praveen\n\nAbstract: \nWe provide numerical eviden
ce that appropriately defined random quantum channels violate additivity o
f Holevo capacity for sending classical information by at least 2^{-182} b
its. The channel achieving this violation acts on a 366 qubit input and gi
ves a 187 qubit output.\n
URL:https://www.tcs.tifr.res.in/web/events/1321
DTSTART;TZID=Asia/Kolkata:20230731T111500
DTEND;TZID=Asia/Kolkata:20230731T121500
LOCATION:A-201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1322
DTSTAMP:20230914T125959Z
SUMMARY:Quantum error correction using Generalized surface codes
DESCRIPTION:Speaker: Pavithran Iyer (Xanadu Quantum Technologies\, Canada)\
n\nAbstract: \nA large-scale quantum computer is envisioned to leverage th
e theoretical guarantees of the fault-tolerant accuracy threshold theorem
to ensure that long computations be carried out reliably\, even in the pre
sence of noise. Quantum Error Correction (QEC) is an integral part of an F
T protocol specifying noise-resilient quantum information storage. Quantum
error-correcting codes encode information in entangled states of many phy
sical qubits. Topological stabilizer codes are a widely studied class of
quantum error-correcting codes wherein the physical qubits are associated
with a tessellation of a surface. The geometric locality of physical qubi
ts in topological codes has both advantages and drawbacks. On the positive
side\, it simplifies the engineering challenges of implementing them on c
ertain hardware platforms. However\, it imposes a strict limitation on the
scaling of the fraction of logical qubits that can be encoded per physica
l qubit. In general\, it is useful to know several parameters of a surface
code to deem its usefulness.\nIn this presentation\, I will introduce a g
eneralized formalism for the study of a specific subclass of topological c
odes called surface codes. Based on findings from [1]\, I will demonstrate
an efficient method to compute the number of logical qubits and the dista
nce of a generalized surface code. Additionally\, I will discuss results f
rom [2]\, showing how to effectively benchmark the performance of generali
zed surface codes under the error model described by quantum erasures. Las
tly\, I will present a tool from [3] that facilitates the analysis of gene
ralized surface codes.\nRelated works:\n[1]: Generalized surface codes and
packing of logical qubits. Nicolas Delfosse\, Pavithran Iyer\, David Poul
in. https://arxiv.org/abs/1606.07116\n[2]: A linear-time benchmarking tool
for generalized surface codes. Nicolas Delfosse\, Pavithran Iyer\, David
Poulin.\nhttps://arxiv.org/abs/1611.04256.\n[3]: SQUAB: A Fast Benchmarkin
g Software for Surface Quantum Computing\nArchitectures. http://quantum-sq
uab.com [1].\n
URL:https://www.tcs.tifr.res.in/web/events/1322
DTSTART;TZID=Asia/Kolkata:20230801T160000
DTEND;TZID=Asia/Kolkata:20230801T170000
LOCATION:in person @ R.No. AG-69 and also via Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1323
DTSTAMP:20230914T125959Z
SUMMARY:Combinatorial Nullstellensatz
DESCRIPTION:Speaker: Rohan Goyal (Chennai Mathematical Institute)\n\nAbstra
ct: \nWe will see a proof of Combinatorial Nullstellensatz and a few of it
s applications.\n
URL:https://www.tcs.tifr.res.in/web/events/1323
DTSTART;TZID=Asia/Kolkata:20230802T160000
DTEND;TZID=Asia/Kolkata:20230802T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1324
DTSTAMP:20230914T125959Z
SUMMARY:STCS Annual Symposium 2023
DESCRIPTION:Speaker: \n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1324
DTSTART;TZID=Asia/Kolkata:20230803T093000
DTEND;TZID=Asia/Kolkata:20230804T183000
LOCATION:AG 69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1325
DTSTAMP:20230914T125959Z
SUMMARY:A sublinear time algorithm for approximately perfect matchings in r
egular graphs
DESCRIPTION:Speaker: Varsha Dani (Golisano College of Computing and Informa
tion Sciences\, Rochester Institute of Technology\, U.S.A.)\n\nAbstract: \
nA breakthrough sequence of papers by Goel\, Kapralov\, and Khanna gave t
he first sublinear-time algorithms for finding large matchings in regular
bipartite graphs. The crown jewel of these is an $O(n \\log n)$ algorith
m to find a perfect matching\, based on the idea of randomly generated au
gmenting paths.\nWe discuss the extent to which the restrictions of being
regular and bipartite can be relaxed. In particular\, we will show that\
, for all d-regular (non-bipartite) graphs\, a matching using a (1-1/(d+1)
) fraction of the vertices can be found in O(n log d) time. At the heart o
f our algorithm is a new method for turning walks that alternate between m
atching and non-matching edges into alternating paths.\n
URL:https://www.tcs.tifr.res.in/web/events/1325
DTSTART;TZID=Asia/Kolkata:20230808T160000
DTEND;TZID=Asia/Kolkata:20230808T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1326
DTSTAMP:20230914T125959Z
SUMMARY:Dynamic Pricing and Matching for Online Marketplaces
DESCRIPTION:Speaker: Sushil Varma (Georgia Tech\, Georgia)\n\nAbstract: \nM
otivated by applications from the gig economy and online marketplaces\, we
study a bipartite matching network under joint pricing and matching contr
ols. The objective is to maximize the long-run average profit and minimize
the delay for the system. In the first part of the talk\, we propose a tw
o-price policy and max-weight matching policy and show that it exhibits a
η1/3 optimality rate when all the arrival rates are scaled by η. We also
demonstrate the advantage of max-weight matching with respect to the numb
er of server and customer types by proving and exploiting state space coll
apse. In the second part of the talk\, we consider the special case of sin
gle customer and server type. The focus is on obtaining the entire distrib
ution of the queue length in heavy traffic. A key observation is that\, un
like a classical queue\, the limiting distribution of a matching queue exh
ibits a phase transition. These results are established by generalizing th
e characteristic function method.\n
URL:https://www.tcs.tifr.res.in/web/events/1326
DTSTART;TZID=Asia/Kolkata:20230810T160000
DTEND;TZID=Asia/Kolkata:20230810T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1327
DTSTAMP:20230914T125959Z
SUMMARY:Some musings on hypercontractivity
DESCRIPTION:Speaker: Eeshan Modak\n\nAbstract: \nHypercontractivity is one
of those interesting ideas that has been discovered/studied independently
by different communities. We will first look at the general definition of
hypercontractivity and two different perspectives of it: one from the info
rmation theory community and the other from the computer science community
. We'll then look at its generalization\, i.e. the Brascamp-Lieb inequal
ity and some of its alternate formulations.\nThis talk is loosely based on
Chandra Nair's tutorial https://www.youtube.com/watch?v=oL1YvAEF450\n
URL:https://www.tcs.tifr.res.in/web/events/1327
DTSTART;TZID=Asia/Kolkata:20230811T160000
DTEND;TZID=Asia/Kolkata:20230811T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1328
DTSTAMP:20230914T125959Z
SUMMARY:Monotone Depth Lower Bounds using Communication Complexity
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya\n\nAbstract: \nWe shall
show that any monotone circuit (with fan-in 2) which determines if a graph
has a perfect matching must have depth $\\Omega(n)$. This shows that effi
cient deterministic parallel algorithms for the perfect matching problem m
ust use negation. To do so\, we shall use tools from communication complex
ity: more specifically\, we shall show that such a circuit will imply a lo
w-cost communication protocol for set-disjointness\, which is known to be
hard.\n
URL:https://www.tcs.tifr.res.in/web/events/1328
DTSTART;TZID=Asia/Kolkata:20230825T160000
DTEND;TZID=Asia/Kolkata:20230825T170000
LOCATION:A-201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1329
DTSTAMP:20230914T125959Z
SUMMARY:Sparsity bounds for factors of sparse polynomials
DESCRIPTION:Speaker: Varun Ramanathan\n\nAbstract: \nThe sparsity of a poly
nomial is defined as the number of monomials in it. We will show that give
n a sparse polynomial f with bounded individual degree\, the sparsity of i
ts factors can be quasi-polynomially bounded by the sparsity of f. This re
sult is from the paper "Deterministic Factorization of Sparse Polynomials
with Bounded Individual Degree" by Bhargava\, Saraf and Volkovich. The pro
of is beautiful and uses techniques from convex geometry.\n
URL:https://www.tcs.tifr.res.in/web/events/1329
DTSTART;TZID=Asia/Kolkata:20230901T160000
DTEND;TZID=Asia/Kolkata:20230901T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1330
DTSTAMP:20230914T125959Z
SUMMARY:Toward Large Kernel Models
DESCRIPTION:Speaker: Parthe Pandit (University of California\, San Diego)\n
\nAbstract: \nRecent studies indicate that kernel machines can often perfo
rm similarly or better than deep neural networks (DNNs) on small datasets.
The interest in kernel machines has been additionally bolstered by the di
scovery of their equivalence to wide neural networks in certain regimes. H
owever\, a key feature of DNNs is their ability to scale the model size an
d training data size independently\, whereas in traditional kernel machine
s model size is tied to data size. Because of this coupling\, scaling kern
el machines to large data has been computationally challenging. In this pa
per\, we provide a way forward for constructing large-scale general kernel
models\, which are a generalization of kernel machines that decouples the
model and data\, allowing training on large datasets. Specifically\, we i
ntroduce EigenPro 3.0\, an algorithm based on projected dual preconditione
d SGD and show scaling to model and data sizes which have not been possibl
e with existing kernel methods.\nParthe is a Simons postdoctoral fellow wi
th the Halıcıoğlu Data Science Institute at UCSD. He obtained his Ph.D.
in ECE from UCLA\, and his undergrad degree in EE from IIT Bombay. He has
been a recipient of the Jack K. Wolf student paper award at ISIT 2019.\n
URL:https://www.tcs.tifr.res.in/web/events/1330
DTSTART;TZID=Asia/Kolkata:20230905T160000
DTEND;TZID=Asia/Kolkata:20230905T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1331
DTSTAMP:20230921T105046Z
SUMMARY:Courcelle's theorem
DESCRIPTION:Speaker: Pranshu Gaba\n\nAbstract: \nGiven a graph\, we are oft
en interested in deciding if it satisfies a certain property. Some example
s of graph properties include connectivity\, bipartiteness\, and planarity
. Courcelle's theorem gives parameterized upper bounds for decidability fo
r a wide class of graph properties. Specifically\, Courcelle's theorem is
a metatheorem that states that if a graph property is definable in monadic
second-order logic on graphs\, then the property is decidable in linear f
ixed-parameter tractable time with the graph treewidth as the parameter. T
his theorem finds many uses in automata theory and model checking. We will
study the theorem and look at some of its applications.\n
URL:https://www.tcs.tifr.res.in/web/events/1331
DTSTART;TZID=Asia/Kolkata:20230908T160000
DTEND;TZID=Asia/Kolkata:20230908T170000
LOCATION:A-201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1332
DTSTAMP:20230914T125959Z
SUMMARY:A Parameterized View on P-matchings
DESCRIPTION:Speaker: Juhi Chaudhary (Ben-Gurion University of the Negev\, I
srael)\n\nAbstract: \nA matching M is a P-matching if the subgraph induced
by the endpoints of the edges of M satisfies property P. For example\, if
the property P is that of being a graph\, being a matching\, being acycli
c\, or being disconnected\, then we obtain the usual matching\, an induced
matching\, an acyclic matching\, and a disconnected matching\, respective
ly. First\, I will survey the latest developments related to P-matchings f
rom the viewpoint of Parameterized Complexity. Then\, I will describe some
results focusing majorly on acyclic matchings and on three algorithmic pa
radigms: approximation hardness\, kernelization lower bounds\, and FPT alg
orithms with respect to various parameters such as treewidth and some belo
w-guarantee parameters. The second part of the talk is based on the two re
cent joint works with Meirav Zehavi\, which appeared in WG’2023.\n \n
URL:https://www.tcs.tifr.res.in/web/events/1332
DTSTART;TZID=Asia/Kolkata:20230912T160000
DTEND;TZID=Asia/Kolkata:20230912T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1333
DTSTAMP:20230920T113426Z
SUMMARY:Quantum Pseudoentanglement
DESCRIPTION:Speaker: Soumik Ghosh (University of Chicago)\n\nAbstract: \nEn
tanglement is a quantum resource\, in some ways analogous to randomness in
classical computation. Inspired by recent work of Gheorghiu and Hoban\, w
e define the notion of "pseudoentanglement''\, a property exhibited by ens
embles of efficiently constructible quantum states which are indistinguish
able from quantum states with maximal entanglement. Our construction relie
s on the notion of quantum pseudorandom states -- first defined by Ji\, Li
u and Song -- which are efficiently constructible states indistinguishable
from (maximally entangled) Haar-random states. Specifically\, we give a c
onstruction of pseudoentangled states with entanglement entropy arbitraril
y close to logn across every cut\, a tight bound providing an exponential
separation between computational vs information theoretic quantum pseudora
ndomness. We discuss applications of this result to Matrix Product State t
esting\, entanglement distillation\, and the complexity of the AdS/CFT cor
respondence.\n
URL:https://www.tcs.tifr.res.in/web/events/1333
DTSTART;TZID=Asia/Kolkata:20230920T100000
DTEND;TZID=Asia/Kolkata:20230920T110000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1336
DTSTAMP:20230921T031459Z
SUMMARY: Conditional Disclosure of Secrets via Non-Linear Reconstruction
DESCRIPTION:Speaker: Ratnakar Medepalli (TIFR)\n\nAbstract: \nConditional D
isclosure of Secrets (CDS) is the problem in which a set of (non-interacti
ng) parties reveal a secret to a receiver if and only if a predicate is tr
ue. We will see results from the Crypto 2017 paper by Liu\, Vaikuntanathan
and Wee in which they demonstrated a connection between CDS schemes and a
class of PIR (Private Information Retrieval) schemes and in the process\,
described the first CDS scheme with sub-exponential communication complex
ity.\n
URL:https://www.tcs.tifr.res.in/web/events/1336
DTSTART;TZID=Asia/Kolkata:20230922T160000
DTEND;TZID=Asia/Kolkata:20230922T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1339
DTSTAMP:20230929T033658Z
SUMMARY:Construction of explicit lossless bipartite expanders
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya (TIFR)\n\nAbstract: \nA
$D\,\\epsilon$ lossless bipartite vertex expander is a $D$-left-regular bi
partite graph such that for each subset $S$ of the left vertices which is
not too large\, the neighborhood of $S$ has size at least $(1-\\epsilon)|S
|$. Lossless expanders are important because of their usage in constructio
n of linear time decodable error correcting codes (Sipser-Spielman codes).
In this talk\, we shall describe the first explicit construction of lossl
ess expanders by Capalbo\, Wigderson et al.\n
URL:https://www.tcs.tifr.res.in/web/events/1339
DTSTART;TZID=Asia/Kolkata:20230929T160000
DTEND;TZID=Asia/Kolkata:20230929T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1338
DTSTAMP:20230927T111049Z
SUMMARY:A Polynomial time Algorithm for the Minimum Generating set Problem
for Groups
DESCRIPTION:Speaker: Dhara Thakkar (IIT Gandhinagar)\n\nAbstract: \nFor a f
inite group G of order n\, a generating set of minimum size is called a mi
nimum generating set of G. A Cayley table for G is a representation of a g
roup to an algorithm. It stores the product of the ith and jth element for
each i\,j ∈ {1\,2\,..\,n}.\nGiven a group G of order n\, by its Cayley
table\, output the size of a minimum generating set problem is known as th
e minimum generating set (MIN-GEN) problem. The MIN-GEN problem admits a t
rivial algorithm that runs in time n^{\\log n+O(1)}. In this talk\, I will
present a polynomial time algorithm that solves the MIN-GEN problem. Our
algorithm also finds one minimum generating set for a given group.\nFinite
groups can also be represented by their generating set as input. Let G \\
leq S_m be a primitive permutation group given by its generating set. We o
btain a quasi-polynomial (in m) time algorithm that outputs the size of a
minimum generating set when G is a primitive permutation group.\nThis talk
is based on the joint work with Bireswar Das\, and Andrea Lucchini.\n
URL:https://www.tcs.tifr.res.in/web/events/1338
DTSTART;TZID=Asia/Kolkata:20231003T160000
DTEND;TZID=Asia/Kolkata:20231003T173000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1341
DTSTAMP:20231005T090319Z
SUMMARY:Is it associative?
DESCRIPTION:Speaker: Shanthanu Suresh Rai (TIFR)\n\nAbstract: \nIs a given
binary operation * on a finite set X associative? We will see a O(n^2) ti
me randomized algorithm for solving this problem (here n = |X|). If time p
ermits\, we will also see how the above algorithm can be extended for chec
king general "read-once" identities.\nReferences:- Jiri Matousek\, Thirty-
three Miniatures: Mathematical and Algorithmic Applications of Linear Alge
bra\, American Mathematical Society\, 2010- Rajagopalan\, Sridhar\; Schulm
an\, Leonard J. (2000). "Verification of Identities". SIAM Journal on Comp
uting. 29 (4): 1155–1163\n
URL:https://www.tcs.tifr.res.in/web/events/1341
DTSTART;TZID=Asia/Kolkata:20231006T154500
DTEND;TZID=Asia/Kolkata:20231006T171500
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1342
DTSTAMP:20231012T044505Z
SUMMARY:Inspiring Trust in Outsourced Computations: From Verifiable Chip Fa
brication to Secure Deep Learning in the Cloud
DESCRIPTION:Speaker: Siddharth Garg (NYU Tandon)\n\nAbstract: \nComputation
s are often outsourced by computationally weak clients to computationally
powerful external entities. Cloud computing is an obvious example of outso
urced computation\; outsourced chip manufacturing to off-shore foundries o
r ``fabs" is another (perhaps less obvious) example. Indeed\, many major
semiconductor design companies have now adopted the so-called "fabless" m
odel. However\, outsourcing raises a fundamental question of trust: how ca
n the client ascertain that the outsourced computations were correctly per
formed? First\, we describe the design of "verifiable ASICs" to address th
e problem of secure chip fabrication at off-shore foundries. Leveraging in
teractive proof (IP) protocols\, we enable untrusted chips to provide run-
time proofs of the correctness of computations they perform. These proofs
are checked by a slower verifier chip fabricated at a trusted foundry. The
proposed approach is the first to defend against arbitrary Trojan misbeha
viors (Trojans refer to malicious modifications of a chip's blueprint by t
he foundry) while providing formal and comprehensive soundness guarantees.
\nNext\, we examine the "MLaaS" setting\, in which both the training and/o
r inference of machine learning models is outsourced to the cloud. MLaaS i
ntroduces both integrity and privacy and integrity risks. ML models can be
maliciously trained\, or provide incorrect outputs during inference. We d
escribe tailored IP protocols for a special class of deep networks that us
e only polynomial activation functions. Finally\, MLaaS also introduces pr
ivacy risk for users since they share their sensitive data with untrusted
cloud applications. Privacy-preserving crypto. methods provide a way out\,
but are exorbitantly expensive. We show how deep network architectures ca
n be tailored to reduce crypto costs by up to two orders of magnitude.\n
URL:https://www.tcs.tifr.res.in/web/events/1342
DTSTART;TZID=Asia/Kolkata:20231013T110000
DTEND;TZID=Asia/Kolkata:20231013T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1337
DTSTAMP:20230925T045659Z
SUMMARY:Harnessing Multiple BMC Engines Together for Efficient Formal Verif
ication
DESCRIPTION:Speaker: Ansuman Banerjee (Indian Statistical Institute\, Kolka
ta)\n\nAbstract: \nIn recent times\, Bounded Model Checking (BMC) engines
are gaining wide prominence and showing great effectiveness in formal veri
fication. Today\, an arsenal of different BMC engines exist\, differing wi
dely in the optimization\, representations and solving mechanisms used to
represent and navigate the underlying state transition system as they look
for property violations. When having a concrete verification task at hand
\, a designer is often confronted with the problem of engine selection\, a
nd more often than not\, has to resort to manually designed selection heur
istics or machine-learned strategies using carefully selected features of
the design. It has often been observed that these different engines have d
ifferent strengths and weaknesses\, depending on the nature of the verific
ation task\, the property and the complexity of the design. This talk will
present a recent work that was set off to examine if combinations of thes
e engines can help to combine the strengths. We talk about a recent work f
rom our group that proposes an approach to create a sequencing of BMC engi
nes that can reach better depths in formal verification\, as opposed to ex
ecuting them alone. Our approach uses machine learning\, specifically\, th
e Multi-Armed Bandit paradigm of Reinforcement Learning\, to predict the b
est-performing BMC engine for a given unrolling depth of the underlying de
sign transition system. We present an evaluation of our approach to show t
hat our proposed approach outperforms state-of-the-art BMC engines in term
s of the depth reached or time taken to deduce a property violation on the
Hardware Model Checking Competition (HWMCC) benchmarks.\n
URL:https://www.tcs.tifr.res.in/web/events/1337
DTSTART;TZID=Asia/Kolkata:20231017T160000
DTEND;TZID=Asia/Kolkata:20231017T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1345
DTSTAMP:20231016T052415Z
SUMMARY:Topology of Omics Data
DESCRIPTION:Speaker: Laxmi Parida (IBM\, NYU Tandon)\n\nAbstract: \nMany pr
oblems in omics are combinatorial in nature\, i.e.\, the interrelationship
amongst the entities is at par\, if not more important\, than the value o
f the entity themselves. Graphs are the most commonly used mathematical ob
ject to model such relationships. However\, often it is important to captu
re higher order relationships as well. Topological data analysis provides
a natural basis to model such interactions and the use of Logic enables ex
traction of signal patterns as logical expressions (or hypothesis) from no
isy data. I will discuss a few applications of such models.\n
URL:https://www.tcs.tifr.res.in/web/events/1345
DTSTART;TZID=Asia/Kolkata:20231020T160000
DTEND;TZID=Asia/Kolkata:20231020T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1346
DTSTAMP:20231020T044050Z
SUMMARY:Counting Markov equivalence classes with the same skeleton
DESCRIPTION:Speaker: Vidya Sagar Sharma (TIFR)\n\nAbstract: \nWe study a di
rected acyclic graphical model\, also known as a Bayesian network. In this
model\, a directed acyclic graph (DAG) is used to represent conditional d
ependence between random variables. Two DAGs are said to be Markov equival
ent if both represent the same set of conditional independence between the
random variables. Verma and Pearl (1990) show that two DAGs are Markov eq
uivalent if\, and only if\, both have the same skeleton (underlying undire
cted graph) and the same v-structures (induced subgraph of the form $a\\ri
ghtarrow b \\leftarrow c$). Markov equivalent DAGs belong to the same Mark
ov equivalent class (MEC). A graphical representation of an MEC is the uni
on of the DAGs it contains.An interesting problem related to the study of
MECs is: for an input undirected graph $G$\, count the MECs with skeleton
$G$. This problem has both graphical and statistical significance. In this
talk\, we will discuss a polynomial algorithm that solves the problem whe
n the skeleton is a tree graph. We will also explore a fixed parameter tra
ctable algorithm to solve the problem.\n
URL:https://www.tcs.tifr.res.in/web/events/1346
DTSTART;TZID=Asia/Kolkata:20231020T180000
DTEND;TZID=Asia/Kolkata:20231020T190000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1349
DTSTAMP:20231025T111829Z
SUMMARY:Counting Markov equivalence classes with the same skeleton
DESCRIPTION:Speaker: Vidya Sagar Sharma (TIFR)\n\nAbstract: \nWe study a di
rected acyclic graphical model\, also known as a Bayesian network. In this
model\, a directed acyclic graph (DAG) is used to represent conditional d
ependence between random variables. Two DAGs are said to be Markov equival
ent if both represent the same set of conditional independence between the
random variables. Verma and Pearl (1990) show that two DAGs are Markov eq
uivalent if\, and only if\, both have the same skeleton (underlying undire
cted graph) and the same v-structures (induced subgraph of the form $a\\ri
ghtarrow b \\leftarrow c$). Markov equivalent DAGs belong to the same Mark
ov equivalent class (MEC). A graphical representation of an MEC is the uni
on of the DAGs it contains.An interesting problem related to the study of
MECs is: for an input undirected graph $G$\, count the MECs with skeleton
$G$. This problem has both graphical and statistical significance. This is
a continuation of last week's talk. In last week's talk\, we discussed a
polynomial time algorithm that solves the problem when the skeleton is a t
ree graph. In this talk\, we will explore a fixed parameter tractable algo
rithm to solve the problem. We will recall the necessary prerequisites fro
m the previous talk.\n
URL:https://www.tcs.tifr.res.in/web/events/1349
DTSTART;TZID=Asia/Kolkata:20231027T170000
DTEND;TZID=Asia/Kolkata:20231027T183000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1351
DTSTAMP:20231031T013539Z
SUMMARY:Statistically Near-Optimal Hypothesis Selection
DESCRIPTION:Speaker: Klim Efremenko (Ben-Gurion University of the Negev)\n\
nAbstract: \nHypothesis Selection is a fundamental distribution learning
problem where given a comparator-class Q={q_1\,...\, q_n} of distribution
s and a sampling access to an unknown target distribution p\, the goal is
to output a distribution q such that tv(p\,q) is close to \\opt\, where $\
\text{opt} = \\min_i\\{\\text{tv}(p\,q_i)\\}$ and $\\text{tv}(\\cdot\, \\c
dot)$ denotes the total-variation distance. Despite the fact that this pro
blem has been studied since the 19th century\, its complexity in terms of
basic resources\, such as a number of samples and approximation guarantees
\, remains unsettled. This is in stark contrast with other (younger) learn
ing settings\, such as PAC learning\, for which these complexities are wel
l understood.\nWe derive an optimal 2-approximation learning strategy for
the Hypothesis Selection problem with a (nearly) optimal sample complexit
y of~$\\tilde O(\\log n/\\epsilon^2)$. This is the first algorithm that si
multaneously achieves the best approximation factor and sample complexity:
previously\, Bousquet\, Kane\, and Moran ({\\it COLT `19}) gave a learner
achieving the optimal $2$-approximation\, but with an exponentially worse
sample complexity of $\\tilde O(\\sqrt{n}/\\epsilon^{2.5})$\, and Yatraco
s~({\\it Annals of Statistics `85}) gave a learner with optimal sample com
plexity of $O(\\log n /\\epsilon^2)$ but with a sub-optimal approximation
factor of $3$.\n
URL:https://www.tcs.tifr.res.in/web/events/1351
DTSTART;TZID=Asia/Kolkata:20231031T160000
DTEND;TZID=Asia/Kolkata:20231031T173000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1350
DTSTAMP:20231027T065509Z
SUMMARY:SGD from the Lens of Markov process: An Algorithmic Stability Persp
ective
DESCRIPTION:Speaker: Anant Raj (University of Illinois at Urbana-Champaign)
\n\nAbstract: \nIn this talk\, we delve into the intricate relationships b
etween heavy-tailed distributions\, generalization error\, and algorithmic
stability in the realm of noisy stochastic gradient descent. Recent resea
rch has illustrated the emergence of heavy tails in stochastic optimizatio
n and their intriguing links to generalization error. However\, these stud
ies often relied on challenging topological and statistical assumptions. E
mpirical evidence has further challenged existing theory\, suggesting that
the relationship between heavy tails and generalization is not always mon
otonic. In response\, we introduce novel insights\, exploring the relation
ship between tail behavior and generalization properties through the lens
of algorithmic stability. Our analysis reveals that the stability of stoch
astic gradient descent (SGD) varies based on how we measure it\, leading t
o interesting conclusions about its behavior.Expanding upon these findings
\, we extend the scope to a broader class of objective functions\, includi
ng non-convex ones. Leveraging Wasserstein stability bounds for heavy-tail
ed stochastic processes\, our research sheds light on the non-monotonic co
nnection between generalization error and heavy tails\, offering a more co
mprehensive perspective.Additionally\, we introduce a unified approach for
proving Wasserstein stability bounds in stochastic optimization\, emphasi
zing time-uniform stability and its role in various scenarios\, including
convex and non-convex losses. Our approach is versatile and applicable to
popular optimizers\, highlighting the importance of ergodicity.\n
URL:https://www.tcs.tifr.res.in/web/events/1350
DTSTART;TZID=Asia/Kolkata:20231101T203000
DTEND;TZID=Asia/Kolkata:20231101T220000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1354
DTSTAMP:20231103T054543Z
SUMMARY:The Optimal Approximation Factor in the Hypothesis Selection Proble
m
DESCRIPTION:Speaker: Eeshan Modak (TIFR)\n\nAbstract: \nConsider the follow
ing problem: Given a collection of probability distributions (q_1\,...\,q_
n) and a sample access to an unknown target distribution p\, find which q_
i is closest to p in total variation. Turns out that this problem in gener
al is not tractable. However\, we can output a q_i such that TV(q_i\,p) <=
\\beta OPT + \\epsilon. Here OPT is the TV between p and best candidate i
n our collection.\nIn this talk\, we will see that the approximation facto
r \\beta has to be at least 3.PS: If you attended Klim Efrimenko's talk th
en he cited this result but did not go into the proof.\n
URL:https://www.tcs.tifr.res.in/web/events/1354
DTSTART;TZID=Asia/Kolkata:20231103T160000
DTEND;TZID=Asia/Kolkata:20231103T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1340
DTSTAMP:20231031T062211Z
SUMMARY:Black-Box Identity Testing of Noncommutative Rational Formulas in D
eterministic Quasipolynomial Time
DESCRIPTION:Speaker: Abhranil Chatterjee (Indian Statistical Institute\, Ko
lkata)\n\nAbstract: \nRational Identity Testing (RIT) is the decision prob
lem of determining whether or not a given noncommutative rational formula
computes zero in the free skew field. It admits a deterministic polynomial
-time white-box algorithm [Garg et al.\, 2016\; Ivanyos et al.\, 2018\; Ha
mada and Hirai\, 2021]\, and a randomized polynomial-time black-box algori
thm [Derksen and Makam\, 2017] via singularity testing of linear matrices
over the free skew field.Designing a subexponential-time deterministic RIT
algorithm in black-box is a major open problem in this area. Despite bein
g open for several years\, this question has seen very limited progress. I
n fact\, the only known result in this direction is the construction of a
quasipolynomial-size hitting set for rational formulas of only inversion h
eight two [Arvind et al.\, 2022].In this talk\, I'll present my recent wor
k where we significantly improve the black-box complexity of this problem
and obtain the first quasipolynomial-size hitting set for all rational fo
rmulas of polynomial size. Our construction also yields a quasi-NC RIT al
gorithm in the white-box setting.\n \nJoint work with V. Arvind (IMSc) an
d Partha Mukhopadhyay (CMI).\n
URL:https://www.tcs.tifr.res.in/web/events/1340
DTSTART;TZID=Asia/Kolkata:20231107T160000
DTEND;TZID=Asia/Kolkata:20231107T173000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1356
DTSTAMP:20231109T100844Z
SUMMARY:Constructive Proofs of Concentration Bounds
DESCRIPTION:Speaker: Ashutosh Singh (TIFR)\n\nAbstract: \nWe will see a com
binatorial proof of the Chernoff-Hoeffding bound\, which says that the sum
of independent {0\,1}-valued random variables is highly concentrated arou
nd the expected value. We will also see proof for the case of [0\,1]-value
d random variables. This is based upon the work of Russell Impagliazzo\, V
alentine Kabanets\, Wolfgang Mulzer\, and Natalia Shenkman.\n
URL:https://www.tcs.tifr.res.in/web/events/1356
DTSTART;TZID=Asia/Kolkata:20231110T160000
DTEND;TZID=Asia/Kolkata:20231110T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1343
DTSTAMP:20231109T045017Z
SUMMARY:Improved constructions of large-dimensional corner-free sets
DESCRIPTION:Speaker: Suhail Sherif (LASIGE\, University of Lisbon)\n\nAbstr
act: \nIn this talk we will discuss some famous open problems in additive
combinatorics:- How large can a subset S of [N] be while managing to avoid
k-term arithmetic progressions?- How large can a subset S of [N]^k be whi
le managing to avoid a k-dimensional corner? (If ⋯ represents a 3-term a
rithmetic progression\, then ⠓ represents a 2-dimensional corner. A more
formal definition will be provided in the talk\, or can be found in the p
aper linked below.)We will also discuss multiparty communication complexit
y\, wherein players are given inputs and they want to compute a function o
f those inputs. In the Number-In-Hand model\, each player can see their ow
n input but not the inputs of others. In the Number-On-Forehead model\, ea
ch player can see the other players' inputs but not their own.In 2021 Lini
al and Shraibman utilized a long-known connection which showed that findin
g good k-party Number-On-Forehead communication protocols for the "Exactly
N" function is equivalent to finding large (k-1)-dimensional-corner-free s
ets. Working in the setting when k=3\, they constructed an explicit protoc
ol that matched a 1946 construction of large corner-free sets. Using the c
ommunication complexity point of view\, they improved upon the protocol to
create even larger sets\, giving the first improvement to the "highest-or
der" term since 1946. This was then subsequently improved by Green later t
hat year.In our work we generalize this method to larger dimensions. We cr
eate explicit communication protocols that match a 1961 construction of la
rge k-dimensional-corner-free sets. We then provide an improvement in the
same vein as Linial and Shraibman and Green\, giving the first improvement
to the "highest-order" term since 1961.This is joint work with Lianna Ham
bardzumyan\, Toniann Pitassi\, Morgan Shirley and Adi Shraibman. The paper
can be found at https://arxiv.org/abs/2309.06554.\n
URL:https://www.tcs.tifr.res.in/web/events/1343
DTSTART;TZID=Asia/Kolkata:20231114T160000
DTEND;TZID=Asia/Kolkata:20231114T173000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1353
DTSTAMP:20231103T060410Z
SUMMARY:Communication with Byzantine Users
DESCRIPTION:Speaker: Neha Sangwan (TIFR)\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1353
DTSTART;TZID=Asia/Kolkata:20231116T150000
DTEND;TZID=Asia/Kolkata:20231116T160000
LOCATION:A-201 and Zoom
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1357
DTSTAMP:20231116T105051Z
SUMMARY:The power of regular and permutation branching programs
DESCRIPTION:Speaker: Hari Krishnan P A (TIFR)\n\nAbstract: \nBranching prog
rams are computational models which are similar to finite automata. They a
re particularly interesting because popular pseudorandom generators can fo
ol certain classes of branching programs. In this talk\, we will see a few
restricted classes of branching programs\, i.e.\, regular and permutation
branching programs\, and how effective they are in simulating other branc
hing programs. All the results are taken from here: https://eccc.weizmann
.ac.il/report/2023/102/\n
URL:https://www.tcs.tifr.res.in/web/events/1357
DTSTART;TZID=Asia/Kolkata:20231117T160000
DTEND;TZID=Asia/Kolkata:20231117T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1334
DTSTAMP:20231031T050325Z
SUMMARY:Load Balancing under Data Locality: Extending Mean-Field Framework
to Constrained Large-Scale Systems
DESCRIPTION:Speaker: Debankur Mukherjee (Georgia Institute of Technology\,
Georgia)\n\nAbstract: \nLarge-scale parallel-processing infrastructures su
ch as data centers and cloud networks form the cornerstone of the modern d
igital environment. Central to their efficiency are resource management po
licies\, especially load balancing algorithms (LBAs)\, which are crucial f
or meeting stringent delay requirements of tasks. A contemporary challenge
in designing LBAs for today's data centers is navigating data locality co
nstraints that dictate which tasks are assigned to which servers. These co
nstraints can be naturally modeled as a bipartite graph between servers an
d various task types. Most LBA heuristics lean on the mean-field approxima
tion's accuracy. However\, the non-exchangeability among servers induced b
y the data locality invalidates this mean-field framework\, causing real-w
orld system behaviors to significantly diverge from theoretical prediction
s. From a foundational standpoint\, advancing our understanding in this do
main demands the study of stochastic processes on large graphs\, thus need
ing fundamental advancements in classical analytical tools.\nIn this prese
ntation\, we will delve into recent advancements made in extending the acc
uracy of mean-field approximation for a broad class of graphs. In particul
ar\, we will talk about how to design resource-efficient\, asymptotically
optimal data locality constraints and how the system behavior changes fund
amentally\, depending on whether the above bipartite graph is an expander\
, a spatial graph\, or is inhomogeneous in nature.\n
URL:https://www.tcs.tifr.res.in/web/events/1334
DTSTART;TZID=Asia/Kolkata:20231121T160000
DTEND;TZID=Asia/Kolkata:20231121T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1355
DTSTAMP:20231109T095402Z
SUMMARY:Fourier Growth of Communication Protocols for XOR Functions
DESCRIPTION:Speaker: Makrand Sinha (University of Illinois Urbana-Champaign
)\n\nAbstract: \nThe Fourier growth of a function refers to the growth of
the sum of absolute values of the level-k Fourier coeﬀicients. Bounds on
the Fourier growth\, even for the first few levels\, have important appli
cations in pseudorandomness and quantum-versus-classical separations. Tigh
t bounds on the Fourier growth have been studied for many classes of funct
ions\, including decision trees and parity decision trees.\n \nWe study t
he Fourier growth of functions associated to communication protocols for X
OR functions (functions evaluated on the point-wise XOR of Alice's and Bob
's inputs). If a protocol C computes an XOR function\, then C(x\,y) is a f
unction of x + y. This motivates us to study the XOR-fiber of the communic
ation protocol C\, defined as h(z) := E[C(x\,y)|x + y = z]. These function
s form a powerful class which includes decision trees and parity decision
trees. Proving tight bounds on the Fourier growth of XOR fibers also has a
pplications to the Gap-Hamming problem and improved quantum versus classic
al separations in communication complexity.\n \nIn this talk\, we present
improved Fourier growth bounds for the XOR-fibers of randomized protocols
that communicate d bits. For the first level\, we show a tight O(sqrt{d})
bound. For the second level\, we show an improved O(d^{3/2}) bound. We co
njecture that the optimal bound is O(d.polylog(n)) and leave this as an op
en question.\n \nOur proof relies on viewing the protocol and its Fourier
spectrum as a martingale. One crucial ingredient we use to control the st
ep sizes is a spectral notion of k-wise independence. Loosely speaking\, t
his corresponds to sets such that the k-th moments of the uniform distribu
tion on the set are well-behaved in all directions. We show how imposing s
pectral k-wise independence on Alice's and Bob's sets allows us to prove b
ounds on the level-k Fourier growth of XOR-fibers. We also provide a way o
f adaptively partitioning a large set into a few spectrally k-wise indepen
dent sets.\n \nJoint work with Uma Girish (Princeton University)\, Avisha
y Tal (UC Berkeley) and Kewen Wu (UC Berkeley).\n
URL:https://www.tcs.tifr.res.in/web/events/1355
DTSTART;TZID=Asia/Kolkata:20231122T143000
DTEND;TZID=Asia/Kolkata:20231122T153000
LOCATION:A-269 (DAA Seminar)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1344
DTSTAMP:20231030T132223Z
SUMMARY:Distinct Elements in Streams: An algorithm for the (text) book
DESCRIPTION:Speaker: Kuldeep S. Meel (University of Toronto)\n\nAbstract: \
nGiven a data stream of m elements\, the Distinct Elements problem is to e
stimate the number of distinct elements in the stream. Distinct Elements h
as been a subject of theoretical and empirical investigations over the pas
t four decades resulting in space-optimal algorithms for it. However\, all
the current state-of-the-art algorithms are often difficult to analyze or
impractical.I will present a simple\, intuitive\, sampling-based space-ef
ficient algorithm whose description and the proof are accessible to underg
raduates with a knowledge of basic probability theory.In addition to the s
implicity\, the approach has significant theoretical and practical implica
tions: our approach allowed us to resolve the open problem of (Discrete) K
lee's Measure Problem in the streaming setting and build a state-of-the-ar
t DNF counter in practice.\n
URL:https://www.tcs.tifr.res.in/web/events/1344
DTSTART;TZID=Asia/Kolkata:20231128T100000
DTEND;TZID=Asia/Kolkata:20231128T113000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1366
DTSTAMP:20231130T100300Z
SUMMARY:Robustness and linear contracts
DESCRIPTION:Speaker: Soumyajit Pyne (TIFR)\n\nAbstract: \nWe consider a gam
e theoretic model with two players- the principal and the agent. The princ
ipal knows some actions available to the agent\, but other\, unknown actio
ns may also exist. The principal evaluates contracts according to their wo
rst-case performance\, with respect to the actions that may or may not be
available to the agent. Under very general circumstances\, the unique opti
mal contract is linear.\nRobustness and Linear Contracts\n
URL:https://www.tcs.tifr.res.in/web/events/1366
DTSTART;TZID=Asia/Kolkata:20231201T143000
DTEND;TZID=Asia/Kolkata:20231201T153000
LOCATION:A-238
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1362
DTSTAMP:20231130T091202Z
SUMMARY:Exploring Size Complexity in Decision Trees
DESCRIPTION:Speaker: Yogesh Dahiya (IMSc Chennai)\n\nAbstract: \nDecision t
rees are one of the simplest and most basic models of computation. Given
a computational task\, in the decision tree model(query model) of compu
tation\, the task is computed by adaptively querying the input while str
iving to minimize the number of queries required. While the model is sim
ple\, it still remains a puzzle with many unresolved questions. Many sig
nificant breakthroughs in recent years have been intricately tied to exp
loring various facets of the query model. Discovery of new connections b
etween proof complexity and query complexity and the development of quer
y-to-communication lifting theorems have played a pivotal role in gettin
g new results in proof complexity\, boolean circuit complexity\, and com
munication complexity. In the decision tree model\, there are two natura
l complexity measures of importance: the depth complexity (the worst-cas
e number of queries asked by a query algorithm) and the size complexity
(the space required to store the query algorithm). While significant att
ention has been devoted to the former\, the latter remains relatively
under-explored.In this talk\, we will investigate the relationship between
size complexity and other complexity measures\, and understand the ad
vantages that randomness offers in the context of size complexity. When
the computation task is a search problem\, a nuanced usage of randomness
by Gat and Goldwasser (ECCC-11) led to the beautiful notion of pseudo-d
eterministic mode of computation. Pseudo-deterministic algorithms are ra
ndomized algorithms that solve search problems by almost always providin
g the same canonical solution (per each input). They aim to address the
inherent variability observed in randomized algorithms\, which often pro
duce different correct results across multiple runs. We will explore the
interplay between determinism\, randomness\, and pseudo-determinism con
cerning size complexity in the decision tree model. Additionally\, we wi
ll discuss more generalized variants of decision trees\, where queries a
re allowed to be functions from specific classes rather than ordinary si
ngle-variable queries.\n
URL:https://www.tcs.tifr.res.in/web/events/1362
DTSTART;TZID=Asia/Kolkata:20231201T160000
DTEND;TZID=Asia/Kolkata:20231201T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1359
DTSTAMP:20231130T095622Z
SUMMARY:Towards Computation- and Communication-Efficient Distributed Learni
ng
DESCRIPTION:Speaker: Pranay Sharma (Carnegie Mellon University\, Pittsburgh
\, Pennsylvania)\n\nAbstract: \n\nModern machine learning (ML) systems rel
y on the data collected at the edge devices to power diverse applications
like predictive typing\, personalized recommendations\, and real-time traf
fic updates. However\, data privacy concerns and network bandwidth constra
ints preclude gathering the entire dataset at a central location for furth
er processing. In the past few years\, federated learning (FL) has emerged
as a natural solution to this problem. The edge devices in FL maintain ex
clusive control of their data and in return\, shoulder part of the computa
tional load of the central server. Google and Apple have already deployed
FL to improve GBoard and Siri.\n \nIn this talk\, I will discuss my work
addressing several challenges in FL. Despite extensive research over the p
ast few years\, the underlying optimization problems solved by most work a
re simple minimization. However\, many ML applications\, like GANs\, robus
t learning\, and reinforcement learning\, can be modeled as min-max proble
ms. I will first describe my work solving nonconvex min-max problems in a
federated setting. In addition to achieving state-of-the-art theoretical c
omputation-communication guarantees\, this work interestingly even improve
s the existing centralized methods. Next\, I will also talk about my work
on FL systems solving minimization problems\, where I quantify the impact
of limited device participation\, where only a small fraction of all the d
evices may be available at any time. I will then discuss a reinforcement l
earning problem in a federated setting\, where we prove linear speedup in
the presence of Markov noise\, answering an existing open question. Finall
y\, I will conclude with some future directions I'm excited about and my b
roader research vision.\n
URL:https://www.tcs.tifr.res.in/web/events/1359
DTSTART;TZID=Asia/Kolkata:20231204T110000
DTEND;TZID=Asia/Kolkata:20231204T120000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1358
DTSTAMP:20231205T044029Z
SUMMARY:Fast Numerical Multivariate Multipoint Evaluation
DESCRIPTION:Speaker: Prahladh Harsha (TIFR)\n\nAbstract: \nMultipoint eval
uation is the computational task of evaluating a polynomial given as a lis
t of coefficients at a given set of evaluation inputs. A straightforward a
lgorithm for this problem is to just iteratively evaluate the polynomial a
t each of the inputs. The question of obtaining faster-than-naive (and ide
ally\, close to linear time) algorithms for this problem is a natural and
basic question in computational algebra.\n \nThe classical FFT algorithm
gives such an algorithm for the special case of univariate polynomials and
a well-structured set of evaluation points (say roots of unity). Only as
recently as last year\, was the multivariate version of this problem for a
ll sets of evaluation points resolved for finite fields due to the works o
f Bhargava\, Ghosh\, Guo\, Kumar & Umans.\n \nThe case of infinite fields
(eg\, reals\, rationals) is complicated due to subtleties arising from th
e bit-complexity of the output compelling one to work with either an appro
ximate version of the problem or an exact version where the algorithm is a
llowed to run in time nearly-linear in the output size. Only as recently a
s 2021\, was the univariate version of this problem over infinite fields r
esolved by Moroz.\n \nIn this talk\, we will show how to extend these res
ults to obtain similar nearly-linear time results for the multivariate ver
sion of the problem over infinite fields such as rationals\, reals both in
the approximate and exact setting.\n \n[Joint work with Sumanta Ghosh\,
Simao Herdade\, Mrinal Kumar and Ramprasad Saptharishi]\n
URL:https://www.tcs.tifr.res.in/web/events/1358
DTSTART;TZID=Asia/Kolkata:20231205T160000
DTEND;TZID=Asia/Kolkata:20231205T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1367
DTSTAMP:20231207T030511Z
SUMMARY:Nearly Equitable Allocations Beyond Additivity and Monotonicity
DESCRIPTION:Speaker: Yeshwant Chandrakant Pandit (TIFR)\n\nAbstract: \nEqu
itability (EQ) in fair division requires that items be allocated such that
all agents value the bundle they receive equally. With indivisible items\
, an equitable allocation may not exist\, and hence we instead consider a
meaningful analog\, EQx\, that requires equitability up to any item. EQx a
llocations exist for monotone\, additive valuations. However\, if (1) the
agents' valuations are not additive or (2) the set of indivisible items in
cludes both goods and chores (positively and negatively valued items)\, th
en before the current work it was not known whether EQx allocations exist
or not.\nWe study both the existence and efficient computation of EQx allo
cations. (1) For monotone valuations (not necessarily additive)\, we show
that EQx allocations always exist. Also\, for the large class of weakly we
ll-layered valuations\, EQx allocations can be found in polynomial time. F
urther\, we prove that approximately EQx allocations can be computed effic
iently under general monotone valuations. (2) For non-monotone valuations\
, we show that an EQx allocation may not exist\, even for two agents with
additive valuations. Under some special cases\, however\, we show the exis
tence and efficient computability of EQx allocations. This includes the ca
se of two agents with additive valuations where each item is either a good
or a chore\, and there are no mixed items.\nThe focus of this talk will b
e on the results obtained under monotone valuations. This is joint work wi
th Siddharth Barman(IISC\, Bangalore)\, Umang Bhaskar(TIFR\, Mumbai) and S
oumyajit Pyne(TIFR\, Mumbai).\n
URL:https://www.tcs.tifr.res.in/web/events/1367
DTSTART;TZID=Asia/Kolkata:20231208T160000
DTEND;TZID=Asia/Kolkata:20231208T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1365
DTSTAMP:20231201T042513Z
SUMMARY:Sparsification: Graphs\, CSPs and Codes
DESCRIPTION:Speaker: Madhu Sudan (Harvard John A. Paulson School of Enginee
ring and Applied Sciences)\n\nAbstract: \nA sparsification of a structure\
, with respect to a class of queries\, produces a compressed representatio
n of the structure while answering every query in the class approximately
correctly. The seminal example of sparsification is "graph sparsification
with respect to cut queries"\, due to works of Karger and Benczur and Karg
er from the 1990s\, showing that every graph can be compressed to near-lin
ear size (in the number of nodes in the graph) while approximately capturi
ng the size of every cut in the graph. In 2015 Kogan and Krauthgamer gener
alized the notion of sparsification to all Constraint Satisfaction Problem
s (CSPs) --- here the structure to be compressed is an instance of the CSP
on n variables and a query is an assignment to the n variables\, and the
goal thus is to approximately determine the number of constraints satisfie
d by the queried assignment. Several follow up works gave some non-trivial
CSPs that allow for near linear (in n) sparsifications\, including classi
fication of all binary CSPs (where each constraint applies to two variable
s) that allow such sparsification\, but the general picture seemed wide op
en.In our works we introduce a new class of sparsification problems\, name
ly code sparsification\, where the structure to be preserved is a linear e
rror correcting code\; the query is a message\, and the goal is to compute
the approximate weight of the encoding of the message. We show that this
class of problems gives the right language to abstract the techniques of K
arger and Benczur and Karger --- and indeed all codes can be sparsified to
length nearly linear in the number of message bits. This generalization a
lready resolves some basic questions in CSP sparsification. A further gene
ralization to additive codes over finite abelian groups gives even powerfu
l results and in particular completely classifies the class of symmetric B
oolean CSPs that allow nearly linear sized sparsification. (Prior to our w
ork even the case of sparsification of 3-XOR constraints was open.) A key
question left open by our work is the algorithmic challenge of finding the
sparsification efficiently --- our work only proves the existence of thes
e.Based on joint works with Sanjeev Khanna (U. Penn.) and Aaron (Louie) Pu
tterman (Harvard).\n
URL:https://www.tcs.tifr.res.in/web/events/1365
DTSTART;TZID=Asia/Kolkata:20231211T143000
DTEND;TZID=Asia/Kolkata:20231211T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1335
DTSTAMP:20231130T075535Z
SUMMARY:Optimal Flexible Consensus and its Application to Ethereum
DESCRIPTION:Speaker: Srivatsan Sridhar (Stanford University\, U.S.A.)\n\nAb
stract: \nClassic BFT consensus protocols guarantee safety and liveness fo
r all clients if fewer than one-third of replicas are faulty. However\, in
applications such as high-value payments\, some clients may want to prior
itize safety over liveness. Flexible consensus allows each client to opt f
or safety against more than one-third faulty replicas\, albeit at the expe
nse of liveness under fewer faults. I will present the first construction
that allows the optimal safety-liveness tradeoff for every client simultan
eously. This construction is modular and is realized as an add-on applied
on top of an existing consensus protocol. I adapt this construction to the
existing Ethereum protocol in the form of optimal flexible confirmation r
ules that clients can adopt unilaterally without requiring system-wide cha
nges. I will show results and insights from implementing the confirmation
rules on an Ethereum client.\nThis talk is based on a paper set to appear
at IEEE S&P 2024 and is a joint work with Joachim Neu\, Lei Yang\, and Dav
id Tse.\n
URL:https://www.tcs.tifr.res.in/web/events/1335
DTSTART;TZID=Asia/Kolkata:20231212T160000
DTEND;TZID=Asia/Kolkata:20231212T170000
LOCATION:A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1363
DTSTAMP:20231201T042609Z
SUMMARY:Deciding Conjugacy of a Rational Relation
DESCRIPTION:Speaker: Amaldev Manuel (IIT Goa)\n\nAbstract: \nRational relat
ions are precisely those relations on words that are definable by transduc
ers\, i.e.\, automata with output. Two words are conjugate if they are cyc
lic shifts of each other\, or equivalently\, of the form xy and yx for som
e words x and y. The conjugacy problem for rational relations asks if a gi
ven rational relation is conjugate. We show that the problem is decidable
in exponential time (and polytime for sumfree rational expressions). The p
roof relies on a generalisation of a classic theorem by Lyndon-Schützenbe
rger from word combinatorics. In the talk\, we will discuss some historic
applications of the conjugacy problem\, our particular motivation to study
the problem\, and a brief sketch of the proof. Joint work with C. Aiswary
a (CMI) and Saina Sunny (IIT GOA)A draft of the paper is available
at: https://arxiv.org/abs/2307.06777\n
URL:https://www.tcs.tifr.res.in/web/events/1363
DTSTART;TZID=Asia/Kolkata:20231213T160000
DTEND;TZID=Asia/Kolkata:20231213T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1370
DTSTAMP:20231212T081930Z
SUMMARY:Online Weighted Facility Location
DESCRIPTION:Speaker: Arghya Chakraborty (TIFR)\n\nAbstract: \nThe classic o
nline facility location problem deals with finding the optimal set of faci
lities in an online fashion when demand requests arrive one at a time and
facilities need to be opened to service these requests. In this work\, we
study a variant where each demand request is a pair (x\,w) where x is the
standard location of the demand while w is the corresponding weight of the
request. The cost of servicing request (x\,w) at facility F is w⋅d(x\,F
). For this variant\, given n requests\, we present an online algorithm at
taining a competitive ratio of O(log n) in the secretarial model for the w
eighted requests and show that it is optimal.\nThis is joint work with Pro
f. Rahul Vaze.\n
URL:https://www.tcs.tifr.res.in/web/events/1370
DTSTART;TZID=Asia/Kolkata:20231214T110000
DTEND;TZID=Asia/Kolkata:20231214T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1369
DTSTAMP:20231216T151338Z
SUMMARY:STCS Farewell party for Jaikumar Radhakrishnan (14 Dec 2023)
DESCRIPTION:Speaker: \n\nAbstract: \nProf. Jaikumar Radhakrishnan's farewel
l party on December 14. Please find below the schedule for the event.https
://www.tcs.tifr.res.in/~jaikumar-farewell/\n
URL:https://www.tcs.tifr.res.in/web/events/1369
DTSTART;TZID=Asia/Kolkata:20231214T134500
DTEND;TZID=Asia/Kolkata:20231214T170000
LOCATION:AG-66 (Lecture Theatre)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1364
DTSTAMP:20231205T090534Z
SUMMARY:Identifiability of Product of Experts Models
DESCRIPTION:Speaker: Leonard J. Schulman (Caltech)\n\nAbstract: \nProduct o
f experts (PoE) are layered networks in which the value at each node is an
AND (or product) of the values (possibly negated) at its inputs. These we
re introduced as a neural network architecture that can efficiently learn
to generate high-dimensional data which satisfy many low-dimensional const
raints---thereby allowing each individual expert to perform a simple task.
PoEs have found a variety of applications in learning. More recently\, th
ey have arisen in the theory of causal networks.\n \nWe study the problem
of identifiability of a product of experts model having a layer of binary
latent variables\, and a layer of binary observables that are iid conditi
onal on the latents. The previous best upper bound on the number of observ
ables needed to identify the model was exponential in the number of parame
ters. We show: (a) When the latents are uniformly distributed\, the model
is identifiable with a number of observables equal to the number of parame
ters (and hence best possible). (b) In the more general case of arbitraril
y distributed latents\, the model is identifiable for a number of observab
les that is still linear in the number of parameters (and within a factor
of two of best-possible). The proofs rely on root interlacing phenomena fo
r some special three-term recurrences.\n \nBased on joint work with Spenc
er Gordon\, Manav Kant\, Eric Ma and Andrei Staicu.\n
URL:https://www.tcs.tifr.res.in/web/events/1364
DTSTART;TZID=Asia/Kolkata:20231215T110000
DTEND;TZID=Asia/Kolkata:20231215T120000
LOCATION:AG-80
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1374
DTSTAMP:20231226T083933Z
SUMMARY:Maximal determinant problem for $l_p$ normed spaces
DESCRIPTION:Speaker: Dr. Arun Maiti (Thapar Institute of Engineering & Tech
nology\, Punjab)\n\nAbstract: \nHadamard's maximum determinant problem ask
s for maximas of the determinant on matrices with entries either -1 or 1.\
nIn this talk\, I will introduce a generalisation of this problem for matr
ices with rows having unit $l_p$ norm. The solution to this problem for sm
all orders will be presented. I will further relate it to the problem of f
inding Auerbach bases of $l_p$ spaces. Some new results and conjectures on
this problem will be discussed.\n
URL:https://www.tcs.tifr.res.in/web/events/1374
DTSTART;TZID=Asia/Kolkata:20231227T160000
DTEND;TZID=Asia/Kolkata:20231227T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1368
DTSTAMP:20240102T092014Z
SUMMARY:Efficient Importance Scenario Generation for Optimization with Rare
Events
DESCRIPTION:Speaker: Anand Deo (IIM Bangalore)\n\nAbstract: \nThis talk pro
vides an overview of how one may employ importance sampling effectively as
a tool for solving stochastic optimization formulations incorporating tai
l risk measures such as Conditional Value-at-Risk. Approximating the tail
risk measure by its sample average approximation\, while appealing due to
its simplicity and universality in use\, requires a large number of sample
s to be able to arrive at risk-minimizing decisions with high confidence.
This is primarily due to the rarity with which the relevant tail events ge
t observed in the samples. In simulation\, Importance Sampling is among th
e most prominent methods for substantially reducing the sample requirement
while estimating probabilities of rare events. Can importance sampling be
used for optimization as well? If so\, what are the ingredients required
for making importance sampling an effective tool for optimization formulat
ions involving rare events? We provide an introductory overview of the two
key ingredients in this regard\, namely\, (i) how one may arrive at a cha
nge of measure prescription at every decision\, and (ii) the prominent tec
hniques available for integrating such a prescription within a solution pa
radigm for stochastic optimization formulations.\nThe talk is based on sev
eral joint works with Karthyek Murthy.\n
URL:https://www.tcs.tifr.res.in/web/events/1368
DTSTART;TZID=Asia/Kolkata:20240102T160000
DTEND;TZID=Asia/Kolkata:20240102T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1379
DTSTAMP:20240103T081941Z
SUMMARY:One Shot Non-Catalytic Distributed Purity Distillation
DESCRIPTION:Speaker: Dr. Sayantan Chakraborty (Centre for Quantum Technolog
ies\, National University of Singapore)\n\nAbstract: \nPure states are an
important resource in many quantum information processing protocols. Howev
er\, even making a fixed pure state\, say $\\vert 0 \\rangle$\, in the lab
oratory requires a considerable amount of effort. Often one ends up with a
mixed state $\\rho$ whose classical description is nevertheless known. He
nce it is important to develop protocols that extract a fixed pure state f
rom a known mixed state. In this work\, we study the problem of extracting
a fixed pure state $\\vert 0 \\rangle^{A'} \\vert{0}\\rangle^{B'}$ from a
known pure state $\\rho^{AB}$ distributed between two parties $A$ and $B$
. Here\, $A'$\, $B'$ are subspaces of $A$\, $B$ and the total amount of pu
rity extracted is $\\log |A'| + \\log |B'|$. The parties can borrow local
pure ancilla\, apply local unitary operations and send a message from $A$
to $B$ through a dephasing channel. If local pure ancilla is borrowed\, it
must be subtracted in order to properly account for the purity extracte
d. We obtain the most efficient achievable bounds on one shot distributed
purity extraction\, in terms of the rate of local ancilla borrowed by the
protocol\, while distilling pure qubits at the best known rate. Our protoc
ols borrow little to no local pure ancilla. Our bounds improve upon the ex
isting bounds for this problem in both one shot as well as asymptotic iid
settings. In particular they subsume all the asymptotic iid results of Dev
etak and Krovi-Devetak. In addition\, we derive upper bounds for the rate
of distillation in the one shot setting\, which nearly match our achievabl
e bounds.\n
URL:https://www.tcs.tifr.res.in/web/events/1379
DTSTART;TZID=Asia/Kolkata:20240103T160000
DTEND;TZID=Asia/Kolkata:20240103T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1377
DTSTAMP:20240103T032207Z
SUMMARY:List Decoding of Tanner and Expander Amplified Codes from Distance
Certificates
DESCRIPTION:Speaker: Mr. Shashank Srivastava (Toyota Technological Institut
e at Chicago (TTIC))\n\nAbstract: \nIn the theory of error-correcting code
s\, list decoding is a relaxation of unique decoding useful for tolerating
higher levels of noise. Design of list decoding algorithms for algebraic
codes\, such as Reed-Solomon\, has found numerous applications in error co
rrection\, as well as in complexity theory and pseudorandomnness. However\
, we know of very few techniques for list decoding algorithms when the cod
e may not have such algebraic structure\, such as Tanner codes which are d
efined using sparse expander graphs.\n \nIn this talk\, I will describe h
ow continuous relaxations based on the Sum-of-Squares hierarchy can be use
d to design the first list decoding algorithm for Tanner codes of Sipser-S
pielman [IEEE Trans. Inf. Theory 1996]. The techniques include a novel pro
of of the Johnson bound for arbitrary codes\, distance proofs for pseudoco
dewords\, and correlation rounding for convex hierarchies. I will also dis
cuss extensions to a distance amplification scheme of Alon-Edmonds-Luby [F
OCS 1995].\n \nBased on joint work with Fernando Granha Jeronimo and Madh
ur Tulsiani.\n
URL:https://www.tcs.tifr.res.in/web/events/1377
DTSTART;TZID=Asia/Kolkata:20240104T160000
DTEND;TZID=Asia/Kolkata:20240104T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1376
DTSTAMP:20240103T041211Z
SUMMARY:How powerful are homogeneous algebraic formulas?
DESCRIPTION:Speaker: Prof. Nutan Limaye (IT University of Copenhagen)\n\nAb
stract: \nProving explicit lower bounds on the size of algebraic formulas
is a long-standing open problem in the area of algebraic complexity theory
. Recent results in the area have indicated a strategy for attacking this
question: show that we can convert a general algebraic formula to a homoge
neous algebraic formula with moderate blow-up in size\, and prove strong l
ower bounds against the latter model. In this talk we will discuss the fea
sibility of the above strategy. Homogeneous formula lower bounds. We will
show lower bounds against ‘weighted’ homogeneous formulas of arbitrary
depth. This is the first such lower bound for arbitrary depth formulas. T
his gives a strong indication that lower bounds against homogeneous formul
as may be within reach.Efficient homogenization. We show that any formula
F for a homogeneous polynomial of degree d can be homogenized over fields
of characteristic 0 as long as $d = s^{o(1)}$. Such a result was previousl
y only known when $d = (\\log s)^{1+o(1)}$ (Raz (J. ACM (2013))).Non-commu
tative homogenization. A recent result of Dutta\, Gesmundo\, Ikenmeyer\, J
indal and Lysikov (2022) implies that to homogenize algebraic formulas of
any depth\, it suffices to homogenize non-commutative algebraic formulas o
f depth just 3. We are able to show strong lower bounds against such homog
enization\, suggesting barriers for this approach. The talk is based on a
joint work with Hervé Fournier\, Srikanth Srinivasan\, and Sébastien Tav
enas. \n
URL:https://www.tcs.tifr.res.in/web/events/1376
DTSTART;TZID=Asia/Kolkata:20240105T143000
DTEND;TZID=Asia/Kolkata:20240105T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1380
DTSTAMP:20240104T091221Z
SUMMARY:Deterministic Algorithms for Low-Degree Factors of Constant-Depth C
ircuits
DESCRIPTION:Speaker: Varun Ramanathan (TIFR)\n\nAbstract: \nFor every const
ant d\, we design a subexponential time deterministic algorithm that takes
as input a multivariate polynomial f given as a constant depth algebraic
circuit over the field of rational numbers\, and outputs all irreducible f
actors of f of degree at most d together with their respective multiplicit
ies. Moreover\, if f is a sparse polynomial\, then the algorithm runs in q
uasipolynomial time. Our results are based on a more fine-grained connecti
on between polynomial identity testing (PIT) and polynomial factorization
in the context of constant degree factors and rely on a clean connection b
etween divisibility testing of polynomials and PIT due to Forbes and on su
bexponential time deterministic PIT algorithms for constant depth algebrai
c circuits from the recent work of Limaye\, Srinivasan and Tavenas.\nThis
will be a conference-style short talk of about 20 minutes\, with questions
in the end.\n
URL:https://www.tcs.tifr.res.in/web/events/1380
DTSTART;TZID=Asia/Kolkata:20240105T160000
DTEND;TZID=Asia/Kolkata:20240105T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1371
DTSTAMP:20240109T044521Z
SUMMARY:Of the data\, by the data\, for the data: an algorithmic viewpoint
DESCRIPTION:Speaker: Ashok Vardhan Makkuva (École Polytechnique Fédérale
de Lausanne (EPFL))\n\nAbstract: \nRanging from arts to science\, in disc
iplines traditionally considered as bastions of human ingenuity\, data-dri
ven algorithms have led remarkable breakthroughs in recent years through C
hatGPT (natural languages)\, AlphaGo (game playing)\, and AlphaFold (biolo
gy). With their ever growing prominence and ubiquity\, there is a growing
consensus that the need of the hour is a fundamental understanding of the
success and pitfalls of these algorithms. To this end\, my research adopts
an interdisciplinary approach to design mathematical foundations and prin
cipled algorithms for problems of great practical relevance. While advanci
ng the success frontiers of data-driven methods\, this approach offers a u
nique mathematical lens to study and understand them.\nIn this talk\, I wi
ll present my contributions along these themes in the fields of informatio
n theory\, machine learning\, and optimization. Through our work on KO cod
es\, I will demonstrate how data-driven algorithms can discover state-of-t
he-art codes for wireless communication\, a fundamental problem at the hea
rt of information and coding theory. This research highlights the great po
tential these methods hold for the design of next generation communication
systems. Next\, I will present our novel algorithmic contribution in opti
mal transport where we design an efficient and reliable algorithm to learn
the optimal transport map between two distributions. These key ideas have
broad applications in biology and medicine including cell perturbation an
alysis and drug discovery. Finally\, I will present our ongoing work on de
signing a solid set of theoretical and algorithmic tools to study large la
nguage models (LLMs) and transformers. Despite their impressive performanc
e\, our understanding of these models is still in infancy and my main goal
here is to develop new insights into their inner workings and faster and
efficient training algorithms. I will conclude with my broader research vi
sion in the realm of data science.\n \nAshok is a postdoctoral researcher
at EPFL with Michael Gastpar. He obtained his PhD in ECE from the Univers
ity of Illinois at Urbana-Champaign in August 2022\, with Pramod Viswanath
and Sewoong Oh. He obtained his Masters in ECE with Yihong Wu also from U
IUC in 2017. Earlier he graduated from IIT Bombay with a B.Tech. in EE and
Minors in Mathematics working with Vivek Borkar. His research interests a
re in foundations of data science in topics including machine learning\, i
nformation theory\, optimization\, and statistics. He is a recipient of Be
st Paper Award from ACM MobiHoc 2019. He is also a recipient of several gr
aduate student awards and fellowships including Joan and Lalit Bahl Fellow
ship (twice)\, Sundaram Seshu International Student Fellowship\, finalist
for the Qualcomm Innovation Fellowship 2018.\n
URL:https://www.tcs.tifr.res.in/web/events/1371
DTSTART;TZID=Asia/Kolkata:20240109T160000
DTEND;TZID=Asia/Kolkata:20240109T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1372
DTSTAMP:20240102T061417Z
SUMMARY:The transformation of regular expressions into finite automata: old
and new results
DESCRIPTION:Speaker: Jacques Sakarovitch (IRIF\, CNRS/Paris Cité Universit
y)\n\nAbstract: \nNot many results in Computer Science are recognised to b
e as basic and fundamental as Kleene Theorem. It states the equality of
two sets of objects that we call now languages. A slight change of focus
on this result shows how it is essentially the combination of two familie
s of algorithms: algorithms that transform a finite automaton into a regul
ar expression on one hand and algorithms that build a finite automaton fro
m a regular expression on the other.In this talk\, I shall consider the al
gorithms of the latter family\, a much laboured subject of both theoretica
l and practical interests. I shall present three different constructions\,
classically attributed to Thompson\, Glushkov\, and Brzozowski-Antimirov
respectively\, and their relationships.We shall then see how the extension
of Kleene Theorem beyond languages: to subsets of arbitrary monoids first
and second to subsets with multiplicity\, leads to a modification of the
construction for the first one and to a radical transformation of the proo
f for the third.All recent results were obtained in joint works wi
th Sylvain Lombardy.\n
URL:https://www.tcs.tifr.res.in/web/events/1372
DTSTART;TZID=Asia/Kolkata:20240111T160000
DTEND;TZID=Asia/Kolkata:20240111T173000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1383
DTSTAMP:20240111T083135Z
SUMMARY:On the degree of polynomials computing square roots mod p
DESCRIPTION:Speaker: Shanthanu Suresh Rai (TIFR)\n\nAbstract: \nFor an odd
prime p\, we say that a polynomial f(X) in F_p[X] computes square roots m
od p if for all non-zero perfect squares a\, f(a)^2 = a (mod p).Problem: C
onstruct a low degree polynomial f(X) that compute square roots mod p.It c
an be easily shown that f(X) has degree between p/4 and p/2. When p = 3 mo
d 4\, it is well-known that f(X)=X^{(p+1)/4} computes square roots\, and t
he degree is as low as possible.When p = 1 mod 4\, previously no non-trivi
al lower bound for degree of f(X) were known. In the paper\, the authors s
how that f(X) has degree at least (p-1)/3. The main ingredient in the proo
f is the following general lemma: powers of low degree polynomial cannot h
ave too many consecutive zero coefficients.In the other direction\, the au
thors show that for infinitely many p = 1 mod 4\, the degree of the polyno
mial computing square roots can be (1/2-Ω(1))p.Paper link: https://eccc.
weizmann.ac.il/report/2023/177/downloadAuthors: Kiran Kedlaya\, Swastik Ko
pparty\n
URL:https://www.tcs.tifr.res.in/web/events/1383
DTSTART;TZID=Asia/Kolkata:20240112T143000
DTEND;TZID=Asia/Kolkata:20240112T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1384
DTSTAMP:20240116T043225Z
SUMMARY:On Approximability of Satisfiable CSPs
DESCRIPTION:Speaker: Prof. Amey Bhangale (University of California\, Rivers
ide)\n\nAbstract: \nConstraint Satisfaction Problems (CSPs) are among the
most well-studied problems in Computer Science. The satisfiability problem
for CSP asks whether an instance of CSP has a fully satisfying assignment
\, i.e. an assignment that satisfies all constraints. This problem is know
n to be in class P or is NP-complete by the recently proved Dichotomy Theo
rem of Bulatov and Zhuk. For a given k-ary predicate P:[q]^k−>{0\,1}\, w
here P−1(1) denotes the set of satisfying assignments\, the problem CSP(
P) has every local constraints of the form the predicate P applied to the
ordered set of k variables (or literals).\nA most natural question is to a
sk for the precise threshold α(P) less than 1 for every NP-complete CSP(P
) such that (i) there is a polynomial-time algorithm that finds an assignm
ent satisfying α(P) fraction of the constraints on a satisfiable instance
and (ii) for every \\epsilon greater than 0\, finding an (α(P)+ϵ) satis
fying assignment is NP-hard. It is reasonable to hypothesize that such a t
hreshold exists for every NP-complete CSP(P). This natural question is wid
e open\, though such thresholds are known for some specific predicates (e.
g.\, 7/8 for 3SAT by Hastad). \nThe talk will present recent work initiat
ing a systematic study of this question\, a relevant analytic hypothesis a
nd some progress on it\, and a work of Raghavendra that answers the questi
on on almost-satisfiable instances. The talk will be based on a series of
joint works with Subhash Khot and Dor Minzer.\n
URL:https://www.tcs.tifr.res.in/web/events/1384
DTSTART;TZID=Asia/Kolkata:20240116T160000
DTEND;TZID=Asia/Kolkata:20240116T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1386
DTSTAMP:20240121T034558Z
SUMMARY:On the projected Aubry set of the rate function associated with lar
ge deviations for stochastic approximations
DESCRIPTION:Speaker: Prof. Henrik Hult (Professor in Mathematical Statistic
s)\n\nAbstract: \nWe consider the problem of minimizing an action potentia
l that arises from large deviation theory for stochastic approximations. T
he solutions to the minimizing problem satisfy\, in the sense of a viscosi
ty solution\, a Hamilton-Jacobi equation. From weak KAM theory\, we know t
hat these viscosity solutions are characterised by the projected Aubry set
. The main result of this paper is that\, for a specific rate function cor
responding to a stochastic approximation algorithm\, we prove that the pro
jected Aubry set is equal to the forward limit set to the limit ODE. \n
URL:https://www.tcs.tifr.res.in/web/events/1386
DTSTART;TZID=Asia/Kolkata:20240122T143000
DTEND;TZID=Asia/Kolkata:20240122T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1373
DTSTAMP:20231226T095541Z
SUMMARY:Optimized Decision Making via Active Learning of Stochastic Hamilt
onians
DESCRIPTION:Speaker: Prof. Chandrajit Bajaj (University of Texas at Austin)
\n\nAbstract: \nA Hamiltonian represents the energy of a dynamical system
in phase space with coordinates of position and momentum. The Hamilton’s
equations of motion are obtainable as coupled symplectic differential equ
ations. In this talk I shall show how optimized decision making (action
sequences) can be obtained via a reinforcement learning problem wherein th
e agent interacts with the unknown environment to simultaneously learn a H
amiltonian surrogate and the optimal action sequences using Hamilton dynam
ics\, by invoking the Pontryagin Maximum Principle. We use optimal control
theory to define an optimal control gradient flow\, which guides the rein
forcement learning process of the agent to progressively optimize the Hami
ltonian while simultaneously converging to the optimal action sequence. Ex
tensions to stochastic Hamiltonians leading to stochastic action sequences
and the free-energy principle shall also be discussed.\nThis is joint wor
k with my students Taemin Heo\, Minh Nguyen.\n
URL:https://www.tcs.tifr.res.in/web/events/1373
DTSTART;TZID=Asia/Kolkata:20240123T160000
DTEND;TZID=Asia/Kolkata:20240123T170000
LOCATION:A-201 Seminar Room
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1375
DTSTAMP:20240124T042057Z
SUMMARY:Formal Methods for Software Reliability and Synthesis
DESCRIPTION:Speaker: Dr. Ashish Mishra (Purdue University (CS))\n\nAbstract
: \nBuilding reliable software has been a classical goal in Computer Scien
ce.The most basic premise of my research is derived from this goal. Can we
make programs safe and reliable using formal techniques while making prog
ramming as a discipline more democratic and accessible to the masses?\n
In this talk\, I will begin by highlighting some of these overarching rese
arch interests and directions.I will primarily present two of my recent wo
rks highlighting the effective use of Refinement types and SMT-based tech
niques for the _verification_ and _synthesis_ of programs.(i) The first is
a specification-guided _synthesis_ procedure that uses Hoare-style pre- a
nd post-conditions to express fine-grained effects of potential library co
mponent candidates to drive a bi-directional synthesis search strategy. It
integrates a conflict-driven learning procedure into the synthesis algori
thm that provides a semantic characterization of previously encountered un
successful search paths used to prune possible candidates' space as synthe
sis proceeds.(ii) The second work is a new _Refinement-Type_ system called
_Coverage Type_ which adapts the recent work in Incorrectness Logic to th
e specification and automated verification of test input generators used i
n modern property-based testing systems. Specifications are expressed in t
he language of refinement types\, augmented with coverage types\, types th
at reflect underapproximate constraints on program behavior.I will conclud
e with some of my completed and ongoing works and future research directio
ns.Particularly\, I will discuss three potential paths I am taking:\n(i) O
ur new deductive CBS using _Semantic Similarity Reduction_ based on a nove
l definition of a _Qualified Tree Automata_.\n(ii) Utilizing _underapproxi
mate_ reasoning for deductive synthesis.\n(iii) Applying program synthesis
to _novel domains_ like Robotics and the need for combining Neural and Sy
mbolic program synthesis approaches\, aka _Neurosymbolic program synthesis
_.\nShort Bio:\nAshish Mishra is a Postdoctoral Researcher at Purdue Unive
rsity\, where he works with Professor Suresh Jagannathan in the areas of
Programming Languages\, Program Verification\, and Program Synthes
is.Ashish obtained his Ph.D. from the Indian Institute of Science\, where
he worked under the supervision of Professor Y. N. Srikant.In a
ddition to his work in Computer Science\, Ashish is also interested in ap
plying technology to public policies and solving social problems. He is c
urrently involved with several Indian NGOs such as PARI (People's Archive
for Rural India)\, Mosali (a startup trying to bring women into workforc
e)\, and others that are involved in Media Monitoring and Research.\
n
URL:https://www.tcs.tifr.res.in/web/events/1375
DTSTART;TZID=Asia/Kolkata:20240130T160000
DTEND;TZID=Asia/Kolkata:20240130T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1390
DTSTAMP:20240202T033112Z
SUMMARY:The Devil's Chessboard problem and the Hamming code
DESCRIPTION:Speaker: Varun Ramanathan (TIFR)\n\nAbstract: \nHere's a proble
m. The devil has captured two people and is playing a game with them for t
heir freedom. Person A will be presented with a chessboard with a penny in
each square (64 total)\, with each penny either heads up or tails up rand
omly. The devil will choose a particular square and point it out to Person
A. Person A then chooses a single square\, and flips the penny in that sq
uare. Afterward\, Person A is sent away and Person B is brought forward. B
ased on the new state of the board\, Person B must point out the same squa
re that the devil did in order to win. The two people can devise a strateg
y beforehand\, but cannot communicate once the game starts. How can they w
in?\nHere's another problem. Alice wants to transmit a sequence of bits to
Bob through a communication channel. But Eve has the ability to discreetl
y flip at most a single bit in the message that Alice sends to Bob. Is the
re a way for Alice to add some redundancy in her messages so that Bob ca
n figure out when a bit has been flipped and also correct it?\nThese two p
roblems are connected\, and we will see how.\n
URL:https://www.tcs.tifr.res.in/web/events/1390
DTSTART;TZID=Asia/Kolkata:20240202T160000
DTEND;TZID=Asia/Kolkata:20240202T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1387
DTSTAMP:20240201T032448Z
SUMMARY:Optimal Server Allocation for Flexible Multi-Server Jobs with Conca
ve Speed-up
DESCRIPTION:Speaker: Prof. Arpan Mukhopadhyay (University of Warwick\, UK)\
n\nAbstract: \nThe majority of jobs submitted to modern computing clusters
and data centres are capable of running on a flexible number of computing
cores or servers. We refer to such jobs as flexible multi-server jobs.
Although allocating more servers to a job results in a higher speed-up in
the job's execution\, it reduces the number of servers available to future
jobs. Furthermore\, the speed-up obtained by a job is typically a concave
\, increasing function of the number of servers allocated to it. So a natu
ral question in this setting is: how to optimally allocate servers to jobs
such that average time a job spends in the system is minimised? This is t
he key question that we shall address in this talk by modelling the system
as a loss system where jobs not finding any servers available upon entry
are blocked. We shall discuss server allocation schemes which result in th
e minimum average sojourn time of accepted jobs while ensuring that the bl
ocking probability of jobs vanishes as the system becomes large (i.e.\, al
l jobs are accepted in the limiting system). We shall consider settings wi
th both linear and sub-linear speed-up functions. We shall also consider s
ettings where the jobs have limited and full system access.\nThe talk will
be based on joint works with Samira Ghanbarian (uWaterloo)\, Ravi R. Mazu
mdar (uWaterloo)\, and Fabrice Guillemin (Orange Labs\, France).\n \nShor
t Bio:\nDr. Arpan Mukhopadhyay is currently an Assistant Professor in the
Department of Computer Science at the University of Warwick\, U.K. His res
earch interests include applied probability\, performance analysis of comp
uter and communication networks\, and distributed network algorithms. He h
as received Best Paper Awards at IFIP Performance 2015 and the Internation
al Teletraffic Congress (ITC) 2015. He was also awarded the Rising Scholar
Award at the International Teletraffic Congress 2018 for his contribution
s to mean-field analysis of large heterogeneous networks.\n
URL:https://www.tcs.tifr.res.in/web/events/1387
DTSTART;TZID=Asia/Kolkata:20240205T113000
DTEND;TZID=Asia/Kolkata:20240205T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1385
DTSTAMP:20240125T054532Z
SUMMARY:Quantum Statistical Inference
DESCRIPTION:Speaker: Dr. Samriddha Lahiry (Harvard University\, Cambridge)\
n\nAbstract: \nQuantum state estimation is a fundamental problem in quantu
m information theory with applications in quantum computing and communicat
ion. To determine the state of a quantum system\, researchers often perfor
m measurements on a set of identically prepared quantum states\, which are
indexed by a parameter. These measurements provide information not only a
bout the parameter itself but also about the quantum states. While questio
ns related to optimal quantum measurements can be elegantly formulated in
the language of mathematical statistics\, the underlying non-commutative s
tructure yields inferential results that are distinctly non-trivial compar
ed to their counterparts in classical statistics. Furthermore\, in contras
t to classical models\, where estimates are constructed solely based on me
asurement outcomes\, quantum models introduce an additional layer of compl
exity because the optimal estimator depends on the choice of measurement a
s well. In classical statistics\, a fundamental paradigm involves approxim
ating complex models with simpler ones. One commonly establishes asymptoti
c equivalence between i.i.d. models\, characterized by a local parameter\,
and a Gaussian shift model. This approximation\, known as local asymptoti
c normality (LAN)\, facilitates the construction of an estimator based on
a procedure in the Gaussian model\, offering comparable risk bounds. Notab
ly\, local asymptotic equivalence can be extended to quantum scenarios\, l
inking quantum i.i.d. models with quantum Gaussian models. In this context
\, we obtain optimal estimators in the complex former models based on opti
mal estimators in the simpler latter models.\nShort Bio:\nSamriddha Lahir
y is a postdoctoral fellow at Harvard University in the Department of Stat
istics. He received his Ph.D. in Statistics from Cornell University in 202
2. His research focuses on asymptotic methods in quantum statistical infer
ence and high-dimensional statistics.\n
URL:https://www.tcs.tifr.res.in/web/events/1385
DTSTART;TZID=Asia/Kolkata:20240206T160000
DTEND;TZID=Asia/Kolkata:20240206T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1392
DTSTAMP:20240209T052522Z
SUMMARY:Wyner's common Information
DESCRIPTION:Speaker: Malhar Ajit Managoli (TIFR)\n\nAbstract: \nGiven two
dependent random variables X and Y\, it is interesting to ask how much inf
ormation is present in both of them. Mutual information is the typical mea
sure of "common" information\; however\, it is not useful in all situation
s. Wyner's common information is another such measure which shows up in ma
ny situations.In this talk we will see the the definition\, some propertie
s\, and two operational interpretations of Wyner's common information.\n
URL:https://www.tcs.tifr.res.in/web/events/1392
DTSTART;TZID=Asia/Kolkata:20240209T160000
DTEND;TZID=Asia/Kolkata:20240209T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1388
DTSTAMP:20240125T064217Z
SUMMARY:New Lower Bounds for Set-Multilinear Branching Programs
DESCRIPTION:Speaker: Mr. Deepanshu Kush (University of Toronto)\n\nAbstract
: \nIn this talk\, we will discuss new lower bounds for the model of sum
of ordered set-multilinear algebraic branching programs. The significance
of these lower bounds is underscored by the recent work of Bhargav\, Dwive
di\, and Saxena (2023)\, which showed that super-polynomial lower bounds a
gainst this model -- for a set-multilinear polynomial of sufficiently low
degree -- would imply super-polynomial lower bounds against general ABPs\
, thereby resolving Valiant's longstanding conjecture that the permanent p
olynomial cannot be computed efficiently by ABPs. We will discuss our new
results which "almost" meet this low-degree demand. This is joint work wit
h Prerona Chatterjee\, Shubhangi Saraf\, and Amir Shpilka. \nShort Bio:\n
Deepanshu Kush is a fourth year PhD student in Computer Science at the Uni
versity of Toronto\, where he is advised by Shubhangi Saraf. His research
interests are broadly in theoretical computer science and related areas of
math with a focus on computational complexity theory.\n
URL:https://www.tcs.tifr.res.in/web/events/1388
DTSTART;TZID=Asia/Kolkata:20240212T113000
DTEND;TZID=Asia/Kolkata:20240212T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1381
DTSTAMP:20240129T063348Z
SUMMARY:Competing Bandits in Non-Stationary Matching Markets
DESCRIPTION:Speaker: Prof. Avishek Ghosh (Indian Institute of Technology\,
Bombay (IITB))\n\nAbstract: \nUnderstanding complex dynamics of two-sided
online matching markets\, where the demand-side agents compete to match wi
th the supply-side (arms)\, has recently received substantial interest. To
that end\, in this paper\, we introduce the framework of decentralized tw
o-sided matching market under non stationary (dynamic) environments. We ad
here to the serial dictatorship setting\, where the demand-side agents hav
e unknown and different preferences over the supply-side (arms)\, but the
arms have fixed and known preference over the agents. We propose and analy
ze an asynchronous and decentralized learning algorithm\, namely Non-Stati
onary Competing Bandits (NSCB)\, where the agents play (restrictive) suc
cessive elimination type learning algorithms to learn their preference ove
r the arms. The complexity in understanding such a system stems from the f
act that the competing bandits choose their actions in an asynchronous fas
hion\, and the lower ranked agents only get to learn from a set of arms\,
not dominated by the higher ranked agents\, which leads to forced explorat
ion. With carefully defined complexity parameters\, we characterize this f
orced exploration and obtain sub-linear (logarithmic) regret of NSCB. Furt
hermore\, we validate our theoretical findings via experiments.\nShort Bio
:\nAvishek Ghosh (Ph.D UC Berkeley\, 2021) is an Assistant Professor at th
e department of Systems and Control Engg. and The Centre for Machine Intel
ligence and Data Science at IIT Bombay. Previously\, he was an HDSI (Data
Science) Post-doctoral fellow at the University of California\, San Diego.
Prior to this\, he completed my PhD from the Electrical Engg. and Compute
r Sciences (EECS) department of UC Berkeley\, advised by Prof. Kannan Ramc
handran and Prof. Aditya Guntuboyina. His research interests are broadly i
n Theoretical Machine Learning\, including Federated Learning and multi-ag
ent Reinforcement/Bandit Learning. In particular\, Avishek is interested i
n theoretically understanding challenges in multi-agent systems\, and comp
etition/collaboration across agents. Before coming to Berkeley\, Avishek c
ompleted his masters degree from Indian Institute of Science (IISc)\, Bang
alore (at the Electrical Communication Engg. Dept) and prior Avishek compl
eted his bachelors degree from Jadavpur University\, in the dept. of Ele
ctronics and Telecommunication Engineering.\n
URL:https://www.tcs.tifr.res.in/web/events/1381
DTSTART;TZID=Asia/Kolkata:20240213T160000
DTEND;TZID=Asia/Kolkata:20240213T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1389
DTSTAMP:20240209T090745Z
SUMMARY:Outlier Oblivious Robust Online Optimization
DESCRIPTION:Speaker: Dr. Adarsh Barik (National University of Singapore (NU
S))\n\nAbstract: \nIn this talk\, I will discuss a robust online convex
optimization framework\, where an adversary can introduce outliers by corr
upting loss functions in an arbitrary number of rounds "k"\, unknown to th
e learner. Our focus is on a novel setting allowing unbounded domains and
large gradients for the losses without relying on a Lipschitz assumption.
We introduce a non-convex (invex) robust loss to mitigate the effects of o
utliers and develop a robust variant of the online gradient descent algori
thm by leveraging our robust loss. We establish tight regret guarantees (u
p to constants)\, both dynamic and static\, with respect to the uncorrup
ted rounds and conduct experiments to validate our theory. Furthermore\, w
e present a unified analysis framework for developing online optimization
algorithms for non-convex (invex) losses\, utilizing it to provide regret
bounds for our robust loss\, which may be of independent interest.\nShort
Bio:\nAdarsh Barik is a postdoc at the Institute of Data Science at the Na
tional University of Singapore. He did his PhD in Computer Science at Purd
ue and his undergrad at IIT Madras before that. His research interests are
broadly in Theoretical and computational aspect of Optimization\, Machine
Learning\, Information Theory and High Dimensional Data Analytics.\n
URL:https://www.tcs.tifr.res.in/web/events/1389
DTSTART;TZID=Asia/Kolkata:20240215T111500
DTEND;TZID=Asia/Kolkata:20240215T121500
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1395
DTSTAMP:20240215T090603Z
SUMMARY:The complexity of solving simple stochastic games
DESCRIPTION:Speaker: Pranshu Gaba (TIFR)\n\nAbstract: \nSimple stochastic g
ames (SSG) are zero-sum games played by two players on a directed graph wi
th a designated target vertex. The players take turns moving a token along
the edges of the game graph. The objective of player 1 is to eventually r
each the target vertex\, whereas the objective of player 2 is to never rea
ch the target vertex. The SSG problem is to find the maximum probability w
ith which player 1 wins the game. We will see that this problem is in NP
∩ coNP.We will also see 3 more problems that are polynomial-time equival
ent to SSG\, and thus also in NP ∩ coNP. Showing that any of these probl
ems are in P would be a major breakthrough.You can read the following pape
rs for more details.- The complexity of stochastic games\, Anne Condon (19
92)https://www.sciencedirect.com/science/article/pii/089054019290048K- The
Complexity of Solving Stochastic Games on Graphs\, Daniel Andersson & Pet
er Bro Miltersen (2009) \nhttps://link.springer.com/chapter/10.1007/978-3
-642-10631-6_13\n \n
URL:https://www.tcs.tifr.res.in/web/events/1395
DTSTART;TZID=Asia/Kolkata:20240216T173000
DTEND;TZID=Asia/Kolkata:20240216T183000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1400
DTSTAMP:20240222T081357Z
SUMMARY:Exponential Separation Between Powers of Regular and General Resolu
tion Over Parities
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya (TIFR)\n\nAbstract: \nPr
oving super-polynomial lower bounds on the size of proofs of unsatisfiabil
ity of Boolean formulas using resolution over parities\, is an outstanding
problem that has received a lot of attention after its introduction by Ra
z and Tzamaret. Very recently\, Efremenko\, Garlík and Itsykson proved th
e first exponential lower bounds on the size of ResLin proofs that were ad
ditionally restricted to be bottom-regular. We show that there are formul
as for which such regular ResLin proofs of unsatisfiability continue to ha
ve exponential size even though there exists short proofs of their unsatis
fiability in ordinary\, non-regular resolution. This is the first super-po
lynomial separation between the power of general ResLin and and that of re
gular ResLin for any natural notion of regularity. Our argument\, wh
ile building upon the work of Efremenko et al\, uses additional ideas from
the literature on lifting theorems.\nThis is joint work with Arkadev Cha
ttopadhyay and Pavel Dvorak.\n
URL:https://www.tcs.tifr.res.in/web/events/1400
DTSTART;TZID=Asia/Kolkata:20240223T143000
DTEND;TZID=Asia/Kolkata:20240223T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1391
DTSTAMP:20240220T062209Z
SUMMARY:Bidding Games on Graphs : in Theory and in Practice
DESCRIPTION:Speaker: Dr. Suman Sadhukhan (University of Haifa)\n\nAbstract:
\nIn a two-player zero-sum graph game\, the players move a token througho
ut the game\, producing an infinite play\, which determines the winner of
the game. Bidding games are graph games in which in each turn\, a bidding
(auction) determines which player moves the token: the players have budget
s\, and in each turn\, both players simultaneously submit the bids that do
not exceed the available budgets. The higher bidder moves the token\, and
pays the bid to the lower bidder (called Richman bidding).The standard so
lution concept of interest in bidding games are threshold budgets: the nec
essary and sufficient budget for winning the game.In this survey talk\, on
the theoretical side\, we will explore discrete bidding games\, where the
keyword discrete stands for the bids having a fixed granularity. We obtai
n membership in NP \\cap coNP for solving parity bidding games with expone
ntially succinct representation.This will be followed by our newly propose
d application of bidding games to a decentralized synthesis problem for mu
lti-objective decision making.Here\, synthesized policies express their sc
heduling urgency via bids and a bounded budget guarantees long-run fairnes
s. Moreover\, our proposed solution framework is modular\, in the sense th
at if one objective changes\, we may still combine the policy for the othe
r objective without having the need to recompute it from scratch.If time p
ermits\, we will also walk through the theoretical development of poorman
counter-part of discrete bidding games\, which we study for the first time
\, where instead of handing over the winning bid to the lower bidder\, the
winner of a bid pays to a bank.These works have been in collaboration wit
h Guy Avni\, Kaushik Mallik\, Tobias Meggendorfer\, Josef Tkadlec\, and Đ
orđe Žikelić.\nShort Bio:\nSuman Sadhukhan is currently a postdoctoral
researcher at the University of Haifa. He obtained his PhD in 2022 from In
ria Rennes where he was advised by Nathalie Bertrand\, Nicolas Markey and
Ocan Sankur. He is interested in\, but not restricted to\, Games on Graphs
\, studying different game theoretic models from formal verification and r
eactive synthesis viewpoint. Prior to his PhD training\, he was a student
at Chennai Mathematical Institute where he studied Theoretical Computer Sc
ience and Mathematics.\n
URL:https://www.tcs.tifr.res.in/web/events/1391
DTSTART;TZID=Asia/Kolkata:20240227T160000
DTEND;TZID=Asia/Kolkata:20240227T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1397
DTSTAMP:20240228T090542Z
SUMMARY:Towards Robust and Reliable Machine Learning: Adversaries and Funda
mental Limits
DESCRIPTION:Speaker: Arjun Bhagoji (University of Chicago)\n\nAbstract: \nW
hile ML-based AI systems are increasingly deployed in safety-critical sett
ings\, they continue to remain unreliable under adverse conditions that vi
olate underlying statistical assumptions. In my work\, I aim to (i) unders
tand the conditions under which a lack of reliability can occur and (ii) r
eason rigorously about the limits of robustness\, during both training and
test phases.In the first part of the talk\, I demonstrate the existence o
f strong but stealthy training-time attacks on federated learning\, a rece
nt paradigm in distributed learning. I show how a small number of compromi
sed agents can modify model parameters via optimized updates to ensure des
ired data is misclassified by the global model\, while bypassing custom de
tection methods. Experimentally\, this model poisoning attack leads to a l
ack of reliable prediction on standard datasets.Test-time attacks via adve
rsarial examples\, i.e. imperceptible perturbations to test inputs\, have
sparked an attack-defense arms race. In the second part of the talk\, I st
ep away from this arms race to provide model-agnostic fundamental limits o
n the loss under adversarial input perturbations. The robust loss is shown
to be lower bounded by the optimal transport cost between class-wise dist
ributions using an appropriate adversarial point-wise cost\, the latter of
which can be efficiently computed via a linear program for empirical dist
ributions of interest.To conclude\, I will discuss my ongoing efforts and
future vision towards building continuously reliable and accessible ML sys
tems by accounting for novel attack vectors and new ML paradigms such as g
enerative AI\, as well as developing algorithmic tools to improve performa
nce in data-scarce regimes.\nShort Bio:\nArjun Bhagoji is a Research Scien
tist in the Department of Computer Science at the University of Chicago. H
e obtained his Ph.D. in Electrical and Computer Engineering from Princeton
University\, where he was advised by Prateek Mittal. Before that\, he rec
eived his Dual Degree (B.Tech+M.Tech) in Electrical Engineering at IIT Mad
ras\, where he was advised by Andrew Thangaraj and Pradeep Sarvepalli. Arj
un's research has been recognized with a Spotlight at the NeurIPS 2023 con
ference\, the Siemens FutureMakers Fellowship in Machine Learning (2018-20
19) and the 2018 SEAS Award for Excellence at Princeton University. He was
a 2021 UChicago Rising Star in Data Science\, a finalist for the 2020 Bed
e Liu Best Dissertation Award in Princeton's ECE Department and a finalist
for the 2017 Bell Labs Prize.\n
URL:https://www.tcs.tifr.res.in/web/events/1397
DTSTART;TZID=Asia/Kolkata:20240229T113000
DTEND;TZID=Asia/Kolkata:20240229T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1403
DTSTAMP:20240229T085137Z
SUMMARY:Proofs from X
DESCRIPTION:Speaker: Sahasranand K R (TIFR)\n\nAbstract: \nWe shall endeavo
ur to provide simple and memorable proofs for some mathematical trivia.\n
URL:https://www.tcs.tifr.res.in/web/events/1403
DTSTART;TZID=Asia/Kolkata:20240301T143000
DTEND;TZID=Asia/Kolkata:20240301T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1382
DTSTAMP:20240228T090601Z
SUMMARY:Verification of Concurrent Programs under Release Acquire
DESCRIPTION:Speaker: Krishna S. (Indian Institute of Technology\, Bombay)\n
\nAbstract: \nThis is an overview of some recent work on the verification
of concurrent programs. Traditionally concurrent programs are interpreted
under sequential consistency (SC). Even though SC is very intuitive and ea
sy to use\, modern multiprocessors do not employ SC for performance reas
ons\, and instead use so called "weak memory models". Some of the well k
nown weak memory models in vogue are Intel x-86\, IBM POWER and ARM.
Weak memory features are also incorporated in the development of modern
programming languages such as C11 and Java. This talk is on the verifica
tion of concurrent programs under the release acquire (RA) semantics\, wit
h the main focus being on decidability and complexity questions.\nShort Bi
o:\nKrishna is a faculty member in the department of computer science and
engineering at IIT Bombay. Her areas of research are automata theory\, log
ics\, and their applications to the verification of quantitative as well a
s distributed/concurrent systems.\n
URL:https://www.tcs.tifr.res.in/web/events/1382
DTSTART;TZID=Asia/Kolkata:20240305T160000
DTEND;TZID=Asia/Kolkata:20240305T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1404
DTSTAMP:20240307T103500Z
SUMMARY:A superpolynomial lower bound against weighted homogeneous formulas
DESCRIPTION:Speaker: Varun Ramanathan (TIFR)\n\nAbstract: \nOn the path to
proving lower bounds against general arithmetic/algebraic models of comput
ation\, a longstanding open problem is to construct explicit polynomials t
hat are superpolynomially hard for arithmetic formulas. Even for the poten
tially weaker model of homogeneous formulas\, we do not have such strong l
ower bounds. Recently\, Fournier\, Limaye\, Srinivasan and Tavenas gave a
superpolynomial lower bound against a related model of computation called
weighted homogeneous formulas\, thus getting us a little (or a lot?) close
r to homogeneous formula lower bounds. The proof goes via a relatively sim
ple application of the probabilistic method. We will try and understand th
is proof. There are no prerequisites\, besides the vague notion of mathema
tical maturity.\n
URL:https://www.tcs.tifr.res.in/web/events/1404
DTSTART;TZID=Asia/Kolkata:20240308T143000
DTEND;TZID=Asia/Kolkata:20240308T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1393
DTSTAMP:20240228T090617Z
SUMMARY:Best of Both Worlds in Fair Division
DESCRIPTION:Speaker: Rohit Vaish (Indian Institute of Technology Delhi)\n\n
Abstract: \nEnsuring fairness is a fundamental aspect of human nature. In
times when an increasing amount of decision-making is being handed over to
automated systems\, it is more important than ever to provide solutions w
ith provable\, mathematically rigorous guarantees. The area of fair divisi
on provides a formal theoretical framework to reason about such questions
in the context of resource allocation.\nThis talk will focus on the fair a
llocation of indivisible resources\, which arises in real-world settings s
uch as inheritance division and university course allocation. Traditional
approaches to this problem involve either randomized allocations that are
fair in expectation or deterministic allocations that are approximately fa
ir. I will discuss an algorithmic framework that unifies randomization and
approximation. Specifically\, I will present an algorithm for finding a r
andomized allocation of indivisible goods that is ex-ante envy-free and ex
-post envy-free up to one good. I will also touch upon some open problems
and potential avenues for future research.\nBased on joint work with Haris
Aziz (UNSW Sydney)\, Rupert Freeman (University of Virginia)\, and Nisarg
Shah (University of Toronto). https://pubsonline.informs.org/doi/full/10.
1287/opre.2022.2432\nShort Bio:\nRohit Vaish is an Assistant Professor in
the Department of Computer Science and Engineering and an associate facult
y member at the Yardi School of Artificial Intelligence\, both located at
IIT Delhi. Previously\, he was a postdoctoral researcher at TIFR and Renss
elaer Polytechnic Institute. Before that\, he completed his PhD from IISc.
Rohit's area of research is computational social choice. This field focus
es on studying collective decision-making problems from a computational pe
rspective. More generally\, he enjoys working on problems at the intersect
ion of economics and computer science.\n
URL:https://www.tcs.tifr.res.in/web/events/1393
DTSTART;TZID=Asia/Kolkata:20240312T143000
DTEND;TZID=Asia/Kolkata:20240312T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1405
DTSTAMP:20240315T092242Z
SUMMARY:Fast multivariate multipoint evaluation
DESCRIPTION:Speaker: Shanthanu Suresh Rai (TIFR)\n\nAbstract: \nProblem: W
e are given a multivariate polynomial f(X_0\, X_1\, ...\, X_{m-1}) with co
efficients from F_p and individual degrees atmost (d-1). For N evaluations
points a_1\, a_2\, ...\, a_N\, compute f(a_i) for all i in [1\, N]We will
first consider the simpler univariate version of this problem (evaluate f
(X) at N points). Next\, we will discuss the algorithm for the general mul
tivariate version by Kedlaya and Umans.\nReferences:- https://www.science
direct.com/science/article/pii/S0022000074800292- https://epubs.siam.org/
doi/10.1137/08073408X\n
URL:https://www.tcs.tifr.res.in/web/events/1405
DTSTART;TZID=Asia/Kolkata:20240315T143000
DTEND;TZID=Asia/Kolkata:20240315T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1410
DTSTAMP:20240322T081707Z
SUMMARY:On the robust hypothesis testing problem
DESCRIPTION:Speaker: Eeshan Modak (TIFR)\n\nAbstract: \nConsider the follow
ing problem: Given a set of probability distributions (p_1\,p_2) and a sam
ple access to an unknown target distribution p\, find which hypothesis (p_
1 or p_2) is closer to p in total variation (TV). In general this is not p
ossible. However\, we can output a p_i (from p_1 or p_2) such that TV(p_i\
,p) <= \\beta OPT + \\epsilon. Here OPT is the TV between p and the best c
andidate in our set. We will show a simple test for which \\beta = 3 (whic
h is in fact optimal). Time permitting\, we will also see that if we are a
llowed to output a distribution not from the set (p_1\, p_2)\, then we can
get \\beta = 2.\n
URL:https://www.tcs.tifr.res.in/web/events/1410
DTSTART;TZID=Asia/Kolkata:20240322T143000
DTEND;TZID=Asia/Kolkata:20240322T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1406
DTSTAMP:20240321T050025Z
SUMMARY:Demystifying the border of algebraic models
DESCRIPTION:Speaker: Nitin Saxena (IIT Kanpur)\n\nAbstract: \nBorder (or ap
proximative) complexity of polynomials plays an integral role in GCT appro
ach to P!=NP. This raises an important open question: can a border circuit
be *efficiently* debordered (i.e. convert from approximative to exact)? O
r\, could the approximation involve exponential-precision which may not be
efficiently simulable? Circuits of depth 3 or 4\, are a good testing grou
nd for this question. Kumar (ToCT'20) proved the universal power
of the border of top-fanin-2 depth-3 circuits. We recently solved some of
the related open questions. In this talk we will outline our result:Borde
r of bounded-top-fanin depth-3 circuits is relatively easy-- it can be com
puted by a polynomial-size algebraic branching program (ABP). Our de-borde
ring paradigm has many applications\, especially in identity testing\, low
er bounds\, and factorization.Based on the works with Prateek Dwivedi & Pr
anjal Dutta (CCC 2021) + (FOCS 2021\, invited to SICOMP) + (FOCS 2022) + (
STOC 2024). [https://www.cse.iitk.ac.in/users/nitin/research.html]\nShort
Bio:\nNitin pursued his education at IIT Kanpur and completed his PhD ther
e. He undertook research stints at Princeton University and the National U
niversity of Singapore before joining CWI Amsterdam and later the Universi
ty of Bonn. Currently\, he holds a professorship at IIT Kanpur and serves
as the N.Rama.Rao Chair\, alongside adjunct faculty roles elsewhere. Notab
ly\, he spearheads the "Center for Developing Intelligent Systems" to tack
le governance challenges at scale. In the realm of algebraic complexity\,
Nitin has made significant strides\, addressing prominent open problems th
rough co-authored papers and pioneering algorithmic-algebraic techniques.
Active within the academic community\, Nitin engages in various committees
and policy-making endeavors. Recognized for his contributions\, he has re
ceived numerous accolades\, including the Gödel Prize\, Fulkerson Prize\,
and Shanti Swarup Bhatnagar Prize.\n
URL:https://www.tcs.tifr.res.in/web/events/1406
DTSTART;TZID=Asia/Kolkata:20240323T160000
DTEND;TZID=Asia/Kolkata:20240323T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1402
DTSTAMP:20240315T044005Z
SUMMARY:Stochastic Gradient Descent using Zero-Order Estimators with Reduce
d Estimation Bias
DESCRIPTION:Speaker: Shalabh Bhatnagar (Indian Institute of Science\, Banga
lore)\n\nAbstract: \nWe present a new family of generalized simultaneous p
erturbation stochastic gradient estimators that estimate the gradient of a
n objective function using noisy function measurements\, but where the num
ber of function measurements and the form of the gradient estimator is gui
ded by the desired estimator bias. Estimators with more function measureme
nts are seen to result in lower bias. We sketch an asymptotic convergence
argument under a constant gradient estimation parameter to the attractor o
f a limiting differential inclusion as well as provide a finite time bound
on the mean-squared error under certain conditions. \nShort Bio:\nShalab
h Bhatnagar received a B.Sc(Hons) in Physics (Delhi University\, 1988)\,
followed by a Master's and PhD in Electrical Engineering from IISc (1992/9
7). He was a Research Associate at the University of Maryland\, College P
ark (1997-2000) and the Vrije Universiteit\, Amsterdam (2000-01). He is w
ith the CSA Department of IISc since December 2001 and is currently a Sen
ior Professor. His interests are in stochastic approximation algorithms\,
reinforcement learning and stochastic optimization. He is a Fellow of all
major science and engineering academies in India and is a J.C.Bose Nationa
l Fellow.\n
URL:https://www.tcs.tifr.res.in/web/events/1402
DTSTART;TZID=Asia/Kolkata:20240402T160000
DTEND;TZID=Asia/Kolkata:20240402T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1412
DTSTAMP:20240403T092106Z
SUMMARY:The complexity of contracts [Roughgarden et. al.]
DESCRIPTION:Speaker: Soumyajit Pyne (TIFR)\n\nAbstract: \nThe paper focuses
on principal-agent settings with succinctly described outcome space (expo
nentially large). The goal of the principal is to design a contract that m
aximizes its expected payoff. The authors show that for n > 2 actions this
problem turns out to be NP-hard. For settings with a constant number of a
ctions\, the authors present a fully polynomial-time approximation scheme
(FPTAS) for the separation oracle of the dual of the problem of minimizing
the principal's payment to the agent\, and use this subroutine to efficie
ntly compute a delta-incentive-compatible (delta-IC) contract whose expect
ed payoff matches or surpasses that of the optimal IC contract.With an arb
itrary number of actions\, the authors prove that the problem is hard to a
pproximate within any constant c. This inapproximability result holds even
for delta-IC contracts where delta is a sufficiently rapidly-decaying fun
ction of c. On the positive side\, the authors show that simple linear del
ta-IC contracts with constant delta are sufficient to achieve a constant-f
actor approximation of the "first-best" (full-welfare-extracting) solution
\, and that such a contract can be computed in polynomial time.\nHere is t
he link to the paper : https://arxiv.org/abs/2002.12034\n
URL:https://www.tcs.tifr.res.in/web/events/1412
DTSTART;TZID=Asia/Kolkata:20240405T160000
DTEND;TZID=Asia/Kolkata:20240405T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1394
DTSTAMP:20240403T042706Z
SUMMARY:Capacitated Facility Location with Outliers
DESCRIPTION:Speaker: Naveen Garg (Indian Institute of Technology Delhi)\n\n
Abstract: \nIn this talk I will present a simple local search algorithm fo
r capacitated facility location when facility costs are uniform. Our algor
ithm is a 3.732-approximation and improves the 4-approximation of Kao. \nI
n the second part of the talk\, I will consider the setting when we are pe
rmitted not to serve L clients (outliers). We extend the local search alg
orithm to obtain the first constant factor approximation for this problem.
Our local search algorithm requires only 2 operations and is a 6.372-appr
oximation.\nShort Bio:\nNaveen Garg is the Usha Hasteer Professor of Compu
ter Science at the Indian Institute of Technology Delhi. He did his B.Tech
. and Ph.D. in Computer Science from IIT Delhi\, was a postdoctoral resear
cher at the Max-Planck-Institut fur Informatik\, Germany and since 1998 he
has been a faculty member at IIT Delhi. He is currently on a sabbatical a
t the University of Warwick\, UK as a Royal Society Wolfson visiting profe
ssor.\nNaveen's contributions are primarily in the design and analysis of
approximation algorithms for NP-hard combinatorial optimization problems a
rising in network design\, scheduling\, routing\, facility location etc. H
e is a Fellow of Indian Academy of Science\, and the Indian National Acade
my of Engineering and was awarded the SS Bhatnagar award for Mathematical
Sciences in 2016.\n
URL:https://www.tcs.tifr.res.in/web/events/1394
DTSTART;TZID=Asia/Kolkata:20240409T160000
DTEND;TZID=Asia/Kolkata:20240409T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1413
DTSTAMP:20240412T043502Z
SUMMARY:An introduction to stabilizer codes
DESCRIPTION:Speaker: Ashutosh Shankar (TIFR)\n\nAbstract: \nQuantum error c
orrection has certain challenges that aren't present in the classical case
: phase flips in addition to bit flips\, the no cloning theorem and more.
I'll talk about stabilizer codes\, including surface codes\, which can det
ect and correct these errors.Basic knowledge of qubits and gates would be
handy in following the talk.\n
URL:https://www.tcs.tifr.res.in/web/events/1413
DTSTART;TZID=Asia/Kolkata:20240412T143000
DTEND;TZID=Asia/Kolkata:20240412T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1398
DTSTAMP:20240416T043235Z
SUMMARY:NP-hardness of testing equivalence to sparse polynomials
DESCRIPTION:Speaker: Chandan Saha (Indian Institute of Science\, Bangalore)
\n\nAbstract: \nThe Polynomial Equivalence (PE) problem asks to check if t
wo given multivariate polynomials are equivalent. Polynomials f(x) and g(x
) are equivalent if there is an invertible matrix A such that f = g(Ax). E
quivalent polynomials represent the same function up to a change in the co
ordinate system. Much is unknown about the exact complexity of PE\, which
is an algebraic analog of the graph isomorphism problem. \nA natural vari
ant of PE is the equivalence testing (ET) problem. The ET problem for a cl
ass C of polynomials asks to decide if a given f is equivalent to some g i
n C. Efficient ET algorithms are known for several natural classes of poly
nomials.\nIn the talk\, we will discuss the complexity of deciding if a gi
ven polynomial f is equivalent to some s-sparse polynomial\, where the par
ameter s is also given. A polynomial is s-sparse if it has at most s monom
ials. We will show that this problem is NP-hard. Moreover\, it is also NP-
hard to approximate (within a "small" factor) the smallest s such that f i
s equivalent to some s-sparse polynomial. If time permits\, we will also d
iscuss an NP-hardness result for testing equivalence to low-support polyno
mials. \nBased on joint work with Omkar Baraskar\, Agrim Dewan\, and Pulk
it Sinha.\nShort Bio:\nChandan Saha is a faculty member in the Department
of Computer Science and Automation at the Indian Institute of Science (IIS
c). He did his PhD and MTech at IIT Kanpur and BE at Jadavpur University.
He was a postdoctoral fellow at the Max Planck Institute for Informatics b
efore joining IISc in 2012. His research is primarily on problems in algeb
raic complexity theory.\n
URL:https://www.tcs.tifr.res.in/web/events/1398
DTSTART;TZID=Asia/Kolkata:20240416T160000
DTEND;TZID=Asia/Kolkata:20240416T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1414
DTSTAMP:20240418T101926Z
SUMMARY:Online Convex Optimization with Switching Cost and Delayed Gradient
s
DESCRIPTION:Speaker: Pranab Panda (TIFR)\n\nAbstract: \nThis paper focuses
on Online Convex Optimization (OCO) problem with linear and quadratic swit
ching costs and by the help of an online algorithm OMGD(Online Multiple Gr
adient Descent) for a class of L smooth and μ strongly convex functions\,
in limited information setting (where at time t we choose x_t without kno
wing f_t and our objective is to minimise f(x) + S(x_t-1\, x_t) summed ove
r t. S(.\,.) is the switching cost)\, we can achieve a competitive ratio o
f at most 4(L + 5) + 16(L + 5)/μ for quadratic switching cost.Online conv
ex optimization captures many crucial real world problems like server mana
gement in data centres\, etc. We will also try to look at how the performa
nce of online algorithms changes even when the switching cost changes from
quadratic to linear.\nPaper: https://arxiv.org/abs/2310.11880\n
URL:https://www.tcs.tifr.res.in/web/events/1414
DTSTART;TZID=Asia/Kolkata:20240419T160000
DTEND;TZID=Asia/Kolkata:20240419T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1399
DTSTAMP:20240405T070007Z
SUMMARY:Approximate Nearest Neighbor Search algorithms for web-scale search
and recommendation
DESCRIPTION:Speaker: Ravishankar Krishnaswamy (Microsoft Research Lab – I
ndia)\n\nAbstract: \n\nWeb-scale search and recommendation scenarios incre
asingly use Approximate Nearest Neighbor Search (ANNS) algorithms to index
and retrieve objects based on the similarity of their learnt representati
ons in a geometric space. Since these scenarios often span billions or tri
llions of objects\, efficient and scalable ANNS algorithms are critical to
making these systems practical.\nIn this talk we discuss some recent empi
rical progress on this problem. Specifically\, we present DiskANN\, an ANN
S algorithm that can index a billion points and serve queries at latencies
of few milliseconds on a single commodity machine. This represents an ord
er of magnitude more points indexed per machine than previous work. We w
ill also discuss some fundamental open problems in this space in the latte
r half of the talk.\nBased on joint works with Harsha Simhadri\, Sujas J S
ubramanya\, Aditi Singh\, Rohan Kadekodi\, Devvrit\, Shikhar Jaiswal\, Mag
dalen Dobson\, Siddharth Gollapudi\, Neel Karia\, Varun Sivashankar\, and
Varun Suriyanarayana.\nShort Bio:\nHe is a principal researcher at Microso
ft Research India. His PhD was completed at Carnegie Mellon University in
2012. From 2012-2014\, a Simons Postdoctoral Fellowship was held by him at
the CS Department in Princeton University. Long\, long ago\, he was an un
dergrad at IIT Madras. Broad interest in basic problems in algorithms and
optimization characterizes his work. Lately\, work on the design of very l
arge-scale (billions of vectors) approximate nearest neighbor search (ANNS
) has been undertaken as part of the DiskANN project. Also\, he spends his
time thinking about basic problems in online and approximation algorithms
\, especially for graph-theoretic and clustering problems.\n
URL:https://www.tcs.tifr.res.in/web/events/1399
DTSTART;TZID=Asia/Kolkata:20240423T160000
DTEND;TZID=Asia/Kolkata:20240423T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1415
DTSTAMP:20240423T101531Z
SUMMARY:On the Nystrom Approximation for Preconditioning in Kernel Machines
DESCRIPTION:Speaker: Parthe Pandit (IIT Bombay)\n\nAbstract: \nKernel metho
ds are a popular class of nonlinear predictive models in machine learning.
Scalable algorithms for learning kernel models need to be iterative in na
ture\, but convergence can be slow due to poor conditioning. Spectral prec
onditioning is an important tool to speed-up the convergence of such itera
tive algorithms for training kernel models. However computing and storing
a spectral preconditioner can be expensive which can lead to large computa
tional and storage overheads\, precluding the application of kernel method
s to problems with large datasets. A Nystrom approximation of the spectral
preconditioner is often cheaper to compute and store\, and has demonstrat
ed success in practical applications. In this paper we analyze the trade-o
ffs of using such an approximated preconditioner. Specifically\, we show t
hat a sample of logarithmic size (as a function of the size of the dataset
) enables the Nystrom-based approximated preconditioner to accelerate grad
ient descent nearly as well as the exact preconditioner\, while also reduc
ing the computational and storage overheads. This is joint work with Amir
hesam Abedsoltan\, Luis Rademacher\, and Mikhail Belkin.\n
URL:https://www.tcs.tifr.res.in/web/events/1415
DTSTART;TZID=Asia/Kolkata:20240424T130000
DTEND;TZID=Asia/Kolkata:20240424T140000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1416
DTSTAMP:20240425T085118Z
SUMMARY:Quantifier Elimination in Presburger Arithmetic
DESCRIPTION:Speaker: Koduri Choudary (TIFR)\n\nAbstract: \nPresburger Arith
metic is the sub-category of First Order Logic\, which deals with quantifi
cation over Integers\, with addition\, order relation\, and multiplication
and divisibility by constants. Our goal is to decide the truth value of s
entence of Presburger Arithmetic by eliminating the quantifiers.Source: De
rek C. Oppen\, A $2^{2^{2^{pn}}}$ Upper Bound on the Complexity of Presbur
ger Arithmetic\, 323-332\, Journal of Computer and System Sciences\, 1978.
https://www.sciencedirect.com/science/article/pii/0022000078900211\n
URL:https://www.tcs.tifr.res.in/web/events/1416
DTSTART;TZID=Asia/Kolkata:20240426T143000
DTEND;TZID=Asia/Kolkata:20240426T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1401
DTSTAMP:20240410T090637Z
SUMMARY:Trading determinism for noncommutativity in Singularity testing
DESCRIPTION:Speaker: Partha Mukhopadhyay (Chennai Mathematical Institute)\n
\nAbstract: \nFinding an efficient deterministic algorithm for symbolic de
terminant identity testing (SDIT) is one of the most important problems in
computational complexity and very little is known about it. Around 2016\,
two independent research groups solved the noncommutative version of the
problem in deterministic polynomial time (Garg-Gurvits-Oliveira-Wigderson
2016\, Ivanyos-Qiao-Subrahmanyam 2017) using very different techniques. In
this talk\, we will discuss a different algorithm for this problem based
on noncommutative polynomial identity testing. Then I will briefly sketch
how this new technique is lifted to approach the partially commutative var
iant of the SDIT and related problems efficiently. The talk is based o
n joint work with V. Arvind\, and Abhranil Chatterjee.\nShort Bio:\nPartha
Mukhopadhyay obtained his PhD from The Institute of Mathematical Sciences
(IMSc\, 2009)\, and then he was a postdoctoral fellow at the Israel Insti
tute of Technology. (Technion\, 2010). Since then\, he is a faculty member
at Chennai Mathematical Institute (CMI). His research interest is mainly
in the design of algorithms for problems which have algebraic flavour. Out
side academics\, he enjoys long distance running\, table tennis\, and pain
ting. \n
URL:https://www.tcs.tifr.res.in/web/events/1401
DTSTART;TZID=Asia/Kolkata:20240430T160000
DTEND;TZID=Asia/Kolkata:20240430T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1420
DTSTAMP:20240502T093307Z
SUMMARY:A generalization of the Coupon Collector Problem
DESCRIPTION:Speaker: Jainam Khakhra (TIFR)\n\nAbstract: \nThe Coupon Collec
tor is a well known problem in Probability Theory and has given rise to se
veral interesting larger problems and generalizations. In this talk\, we d
iscuss the Coupon Collector Problem and a specific generalization of it by
introducing the notion of a Super Coupon which is an s-sized subset of a
universe of n coupons. If time permits we discuss this in the context of R
andom Walks on Graphs. The talk will be based on the paper\, "On a General
ization of the Coupon Collector Problem" by Siva Athreya\, Satyaki Mukherj
ee and Soumendu Sundar Mukherjee.\nLink to paper: https://arxiv.org/pdf/2
304.01145\n
URL:https://www.tcs.tifr.res.in/web/events/1420
DTSTART;TZID=Asia/Kolkata:20240503T143000
DTEND;TZID=Asia/Kolkata:20240503T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1396
DTSTAMP:20240501T045410Z
SUMMARY:Unlikely Friendships: Fruitful Interplay of Mathematical Structures
in Cryptography
DESCRIPTION:Speaker: Shweta Agrawal (Indian Institute of Technology\, Madra
s)\n\nAbstract: \nThe security of cryptographic protocols is based on the
conjectured intractability of some mathematical problem\, typically a sin
gle problem. However\, in some cases\, novel constructions emerge out of
the surprising interplay of seemingly disparate mathematical structures a
nd conjectured hard problems on these. Though unusual\, this cooperation b
etween assumptions\, when it happens\, can lead to progress on important o
pen problems. This sometimes paves the way for subsequent improvements\, w
hich may even eliminate the multiplicity and reduce security to a single a
ssumption.\nIn this talk\, I will give examples from recent work that have
benefited from this interplay\, leveraging diverse mathematical structur
es and assumptions in a way that yields more than a "sum of parts"\, and l
eads to surprising progress in longstanding problems.\nShort Bio:\nDr. Shw
eta Agrawal is a professor at the Computer Science and Engineering departm
ent\, at the Indian Institute of Technology\, Madras. She earned her PhD a
t the University of Texas at Austin\, and did her postdoctoral work at the
University of California\, Los Angeles. Her area of research is cryptogra
phy with a focus on post quantum cryptography. She has won multiple awards
and honours such as the Swarnajayanti award\, the ACM India award for Out
standing Contributions to Computing by a Woman\, a best paper award at Eur
ocrypt\, Google India faculty award\, BY award for excellence in research
and teaching\, invited speaker at prestigious conferences like Asiacrypt a
nd "Women in Mathematics" and program co-chair for the flagship conference
Asiacrypt.\n
URL:https://www.tcs.tifr.res.in/web/events/1396
DTSTART;TZID=Asia/Kolkata:20240507T160000
DTEND;TZID=Asia/Kolkata:20240507T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1422
DTSTAMP:20240509T035847Z
SUMMARY:Online Age-of-Information Scheduling
DESCRIPTION:Speaker: Kumar Saurav (TIFR)\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1422
DTSTART;TZID=Asia/Kolkata:20240509T110000
DTEND;TZID=Asia/Kolkata:20240509T130000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1423
DTSTAMP:20240509T082219Z
SUMMARY:Resolution is NP hard to automate
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya (TIFR)\n\nAbstract: \nA
proof system P is called automatizable if there is an algorithm A that\, g
iven input a statement phi\, runs in time poly(s) and outputs a P-proof of
phi of size at most poly(s)\, where s is the size of the shortest P-proof
of phi. It had been a long-standing open problem to determine if resoluti
on is automatizable. In a breakthrough result\, Atserias and Muller showed
that automating resolution is NP hard. They construct a polynomial time c
omputable transformation A that\, on input a CNF phi\, outputs an unsatisf
iable CNF A(phi) such that:(i) If phi is satisfiable\, A(phi) has polynomi
al sized resolution refutations(ii) If phi is not satisfiable\, A(phi) doe
s not have any subexponential sized resolution refutationWe shall describe
this construction.\n
URL:https://www.tcs.tifr.res.in/web/events/1423
DTSTART;TZID=Asia/Kolkata:20240510T160000
DTEND;TZID=Asia/Kolkata:20240510T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1411
DTSTAMP:20240502T053036Z
SUMMARY:Navigating the Transformative Landscape of Generative AI: Some Stat
istical and Ethical Perspectives of the Large Language Models
DESCRIPTION:Speaker: Swagatam Das (Indian Statistical Institute (ISI)\, Kol
kata)\n\nAbstract: \nIn this talk\, we embark on a journey through the dyn
amic landscape of Generative AI\, delving into both statistical intricacie
s and ethical considerations surrounding Large Language Models (LLMs). We
dissect the statistical modeling techniques employed in generative models\
, shedding light on the emergence of GAN and diffusion models and the pivo
tal role transformers play in sequential data. Amidst the transformative
potential of these models\, we confront the pressing issue of bias and et
hical concerns\, exploring avenues for mitigating algorithmic biases and p
romoting ethical AI practices. The talk will finally touch upon some of th
e open problems that exist in this field. \nShort Bio:\nSwagatam Das\, a
Professor at the Indian Statistical Institute\, Kolkata\, earned his B.E.\
, M.E.\, and Ph.D. degrees from Jadavpur University. Formerly Deputy Direc
tor at TCG CREST\, his research spans deep learning and non-convex optimiz
ation\, with over 400 publications. He co-founded Swarm and Evolutionary C
omputation and serves as Associate Editor for prestigious journals. With 3
2\,000+ Google Scholar citations and an H-index of 84\, he's renowned inte
rnationally\, participating in top conferences like NeurIPS and receiving
accolades such as the 2012 Young Engineer Award and the 2015 Thomson Reute
rs Research Excellence India Citation Award.\n
URL:https://www.tcs.tifr.res.in/web/events/1411
DTSTART;TZID=Asia/Kolkata:20240514T110000
DTEND;TZID=Asia/Kolkata:20240514T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1426
DTSTAMP:20240516T083228Z
SUMMARY:The connection between circuit complexity and first-order logic
DESCRIPTION:Speaker: Pranshu Gaba (TIFR)\n\nAbstract: \nDescriptive complex
ity is a branch of computational complexity where complexity classes are d
escribed using forms of logic\, giving machine-independent characterizatio
ns of the complexity classes. In this talk\, we look at an important resul
t in descriptive complexity:[BIS88\, Theorem 8.1] A class of structures is
in DLOGTIME-uniform AC0 if and only if the class is definable in first-or
der logic with numerical predicates.Here\, DLOGTIME is the class of langua
ges accepted by a deterministic random-access Turing machine in logarithmi
c time.A language is said to be in DLOGTIME-uniform AC0 if it can be recog
nized by a family of constant-depth\, polynomial-size circuits with AND an
d OR gates\, where the circuits can be constructed by a DLOGTIME machine.
The proof involves showing that if a language is in DLOGTIME\, then it is
definable in first-order logic with numerical predicates.Reference:[BIS88]
David A. Mix-Barrington\, Neil Immerman\, and Howard Straubing. 1990. On
uniformity within NC1. J. Comput. Syst. Sci. 41\, 3 (Dec. 1990)\, 274–30
6. https://doi.org/10.1016/0022-0000(90)90022-D\n
URL:https://www.tcs.tifr.res.in/web/events/1426
DTSTART;TZID=Asia/Kolkata:20240517T113000
DTEND;TZID=Asia/Kolkata:20240517T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1428
DTSTAMP:20240522T041810Z
SUMMARY:Local Correction of Linear Functions over the Boolean Cube
DESCRIPTION:Speaker: Madhu Sudan (Harvard John A. Paulson School of Enginee
ring and Applied Sciences)\n\nAbstract: \nSince the late 1980's it has bee
n known that low-degree multivariate polynomials over finite fields form "
locally correctible codes". In other words\, given oracle access to a func
tion f:F^n -> F such there is a degree d polynomial P(X1...Xn) whose evalu
ation agree with f on all but a tiny fraction of points\, and an arbitrary
point b in F^n\, it is possible to determine P(b) with high probability
while querying f in a constant number of points (where the number of queri
es depends on the degree d\, but not the number of variables n).Of course
this result doesn't work for infinite fields since then one has to define
a measure over F^n. But can one hope for a similar result for f:S^n \\to F
for some finite subset S of F? In a previous work with Bafna and Srinivas
an we had shown that any such local decoding result (over say the reals or
rationals) would require roughly Omega-tilde(\\log n) queries (even when
d = 1).In this talk we describe a complementary result that shows that O-t
ilde(\\log n) queries suffice in the linear (d=1) case when S = {0\,1} (ov
er any field F\, and indeed any abelian group G). Working over sets like S
^n provides novel challenges due to the fact that the most handy tool in p
revious works\, namely affine-invariance of the domain\, is no longer avai
lable. Our local reconstruction algorithms rely centrally on the ability t
o construct a small set of nearly balanced vectors in {-1\,1}^n whose span
contains 1^n. Combined with a double dose of hypercontractivity this lead
s to a local algorithm correcting up to 1/4 fraction of errors. Our fina
l result gives a list-decoding algorithm correcting nearly 50% errors. Her
e the key ingredient is the combinatorial bound on the list size which we
prove using a kitchen sink of techniques that show large lists must contai
n many sparse polynomials and then ruling out such possibilities which inv
olves case analysis depending on the group G.\nJoint work with Prashanth A
mireddy\, Amik Raj Behera\, Manaswi Paraashar\, and Srikanth Srinivasan.\n
URL:https://www.tcs.tifr.res.in/web/events/1428
DTSTART;TZID=Asia/Kolkata:20240528T140000
DTEND;TZID=Asia/Kolkata:20240528T150000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1427
DTSTAMP:20240521T095005Z
SUMMARY:Learning in Large and Structured Environments: Algorithms\, Guarant
ees and Applications
DESCRIPTION:Speaker: Sayak Chowdhury (Microsoft Research Lab\, India)\n\nAb
stract: \n\nIn this talk\, I will present some of my work on designing and
analyzing algorithms for learning in large and structured environments\,
where the state and action spaces are huge or even infinite. I will focus
on two main topics: (1) Bayesian optimization for hyperparameter tuning in
large-scale machine learning models\, and (2) Policy optimization for lan
guage models using human feedback. For the first topic\, I will introduce
the Gaussian process optimization framework and design multi-armed-bandi
t algorithms for hyperparameter optimization. I will show sublinear regre
t bounds for the proposed algorithms that depend on the information comple
xity of the objective function to be optimized. Along the way\, I will pr
esent a self-normalized concentration inequality for vector-valued marting
ales of arbitrary\, possibly infinite\, dimension\, and discuss some appli
cations of this concentration bound. For the second topic\, I will talk a
bout the effects of noisy preference data that can negatively impact lang
uage model alignment. I will propose a robust loss function for language
model policy optimization in the presence of random preference flips. I w
ill show that the proposed language model policy is provably tolerant to
noise and characterize its sub-optimality gap as a function of noise rate\
, dimension of the policy parameter\, and sample size. I will also demons
trate the empirical performance of the proposed policy on various tasks\,
such as dialogue generation and sentiment analysis. I will conclude with
some open problems and future directions of research in large scale machin
e learning.\n \nShort Bio:\nSayak Ray Chowdhury is a postdoctoral researc
her at Microsoft Research\, India. Prior to this he was a postdoctoral fel
low at Boston University\, USA. He obtained his PhD from the Dept of ECE\,
Indian Institute of Science\, where he was a recipient of Google PhD fell
owship. His research interests include reinforcement learning\, Bayesian o
ptimization\, multi-armed bandits and differential privacy. Recently\, he
has been working towards mathematical and empirical understandings of lang
uage models. More details about his research can be found here: https://si
tes.google.com/view/sayakraychowdhury/home\n \n
URL:https://www.tcs.tifr.res.in/web/events/1427
DTSTART;TZID=Asia/Kolkata:20240529T160000
DTEND;TZID=Asia/Kolkata:20240529T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1408
DTSTAMP:20240514T040141Z
SUMMARY:Knapsack: Connectedness\, Path\, and Shortest-Path
DESCRIPTION:Speaker: Palash Dey (Indian Institute of Technology\, Kharagpur
)\n\nAbstract: \nWe study the Knapsack problem with graph-theoretic constr
aints. That is\, there exists a graph structure on the input set of items
of Knapsack and the solution also needs to satisfy certain graph theoretic
properties on top of the Knapsack constraints. In particular\, we study C
onnected Knapsack where the solution must be a connected subset of items w
hich has maximum value and satisfies the size constraint of the knapsack.
We show that this problem is strongly NP-complete even for graphs of maxim
um degree four and NP-complete even for star graphs. On the other hand\, w
e develop an algorithm running in time O(2^{tw log tw} · poly (n) min{s^2
\, d^2}) where tw\, s\, d\, n are respectively treewidth of the graph\, th
e size of the knapsack\, the target value of the knapsack\, and the number
of items. We also exhibit a (1 − epsilon) factor approximation algorith
m running in time O(2^{tw log tw} · poly (n\, 1/epsilon)) for every epsil
on > 0. We show similar results for Path Knapsack and Shortest Path Knapsa
ck\, where the solution must also induce a path and shortest path\, respec
tively. Our results suggest that Connected Knapsack is computationally the
hardest\, followed by Path Knapsack and then Shortest Path Knapsack.\nSho
rt Bio:\nProf. Palash Dey is an Assistant Professor in the Department of C
omputer Science and Engineering at the Indian Institute of Technology Khar
agpur. Before joining IIT Kharagpur\, he spent around a year as a visiting
post-doctoral fellow at the School of Technology and Computer Science (ST
CS) at Tata Institute of Fundamental Research (TIFR) Mumbai. Before that\,
he finished his PhD and Master of Engineering from the Department of Comp
uter Science and Automation in 2017 and 2013\, respectively. He received t
he INSPIRE Faculty Award in 2018 and the ACM India Best Dissertation Award
in 2017. He has been a fellow of the West Bengal Academy of Science and T
echnology\, Government of West Bengal\, India\, since 2022 and an INAE You
ng Associate since 2023. He is an ACM India Eminent Speaker for 2024-26. H
is research focuses on Algorithmic Game Theory and Parameterized Algorithm
s.\n
URL:https://www.tcs.tifr.res.in/web/events/1408
DTSTART;TZID=Asia/Kolkata:20240604T160000
DTEND;TZID=Asia/Kolkata:20240604T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1425
DTSTAMP:20240527T102107Z
SUMMARY:Bridging the Theory and Practice of Cryptography
DESCRIPTION:Speaker: Ashrujit Ghoshal (Carnegie Mellon University)\n\nAbstr
act: \nIn the current internet landscape\, cryptography plays a central r
ole in securing communication. We rely on mathematical proofs to ensure s
ecurity of the cryptographic systems that are deployed in practice. Howev
er\, in many cases\, due to issues like efficiency constraints\, there is
a gap between what these deployments need and what we can prove. In this
talk\, I will describe how my research identifies these gaps and makes p
rogress towards bridging these gaps using new theoretical insights and tec
hniques from different areas of computer science like complexity theory\,
algorithms\, combinatorics\, information theory\, etc.More concretely\,
my work contributes towards bridging these gaps in three different ways. F
irst\, I provide exact security analyses of cryptographic systems that hav
e been deployed at scale that did not have such analyses before. With the
exact analyses available\, practitioners can set parameters of the cryptog
raphic system in a way that maximizes efficiency without sacrificing secur
ity. Secondly\, I construct new cryptographic schemes that are better than
existing schemes in terms of efficiency. This work helps make purely theo
retical cryptographic notions practical. Finally\, my work incorporates ne
wer perspectives into the framework of security proofs that captures a mor
e complete picture of the real world. This is in contrast to prior work wh
ere only certain adversarial resources were taken into account. A more com
plete picture of adversarial resources often helps in setting parameters i
n a way that increases efficiency of cryptographic systems.\nShort Bio: As
hrujit Ghoshal is a postdoctoral fellow at Carnegie Mellon University. He
received his PhD from the University of Washington in 2023. His research
focuses on bridging the gap between the theory and practice of cryptogra
phy by developing new theory that characterizes security and efficiency o
f cryptographic systems as precisely as possible. In particular his work
has provided exact security analyses for cryptography that is widely used
in practice e.g.\, standard hash functions like SHA-2 and SHA-3\, TLS\,
etc. His work has also made progress towards making theoretical cryptogra
phic functionalities like private information retrieval more practical by
giving new concretely efficient constructions. These works have led to m
ultiple papers at the two top cryptography conferences- CRYPTO and EUROCR
YPT.\n
URL:https://www.tcs.tifr.res.in/web/events/1425
DTSTART;TZID=Asia/Kolkata:20240605T160000
DTEND;TZID=Asia/Kolkata:20240605T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1421
DTSTAMP:20240521T095047Z
SUMMARY:Lower Bounds for Planar Arithmetic Circuits
DESCRIPTION:Speaker: Ramya C. (The Institute of Mathematical Sciences)\n\nA
bstract: \nArithmetic circuits are a natural computational model for compu
ting multivariate polynomials over a field. Planar arithmetic circuits are
circuits whose underlying graph is planar. The size of a circuit which
is the number of vertices in the underlying graph is a fundamental paramet
er concerning circuits. In this talk\, we will prove a superlinear lower
bound on the size of planar arithmetic circuits computing an explicit bil
inear form. More generally\, we will walk-through the algebraic complexity
of bilinear forms. Furthermore\, Baur and Strassen(1983) showed that all
the first order partial derivatives of a polynomial can be simultaneously
computed with only a constant factor blow-up in size. We observe that an a
nalogous statement does not hold in the case of planar circuits.\nThis tal
k is based on joint work with Pratik Shastri(IMSc).\nShort Bio:\nRamya is
currently a faculty member in the Theoretical Computer Science group at th
e Institute of Mathematical Sciences(IMSc)\, Chennai.Prior to this\, she w
as a faculty member at the Chennai Mathematical Institute(CMI). Before joi
ning CMI\, she was a postdoctoral fellow at STCS\, TIFR\, Mumbai during
2019-2021.She obtained her PhD from IIT Madras in 2019. Her research is ce
ntered around Computational Complexity Theory\, particularly problems with
an algebraic flavor.\n
URL:https://www.tcs.tifr.res.in/web/events/1421
DTSTART;TZID=Asia/Kolkata:20240611T160000
DTEND;TZID=Asia/Kolkata:20240611T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1419
DTSTAMP:20240603T043420Z
SUMMARY:List decoding and Higher Order MDS codes
DESCRIPTION:Speaker: Manik Dhar (Massachusetts Institute of Technology)\n\n
Abstract: \nA [n\,k] code C is a k-dimensional subspace in F_q^n. C is sai
d to be MDS (maximal distance separable) if its hamming weight is n-k+1 wh
ich is the largest possible by the singleton bound. It is also equivalent
to saying that every k minor of the generator matrix of C is invertible. T
his can be rewritten as saying that any two subspaces formed by the column
s of the generator matrix of C intersect as minimally as possible.\nBraken
siek\, Gopi\, and Makam generalize this notion to introduce MDS(l) codes w
here we ask that any l subspace formed by the columns of the generator mat
rix of C intersect as minimally as possible. In a series of works\, they s
how their new notion is equivalent to an alternative notion of higher-orde
r MDS codes introduced by Roth related to list decoding. In particular\, t
he parity check matrix of code being MDS(l) is equivalent to having (avera
ge-case) combinatorial list decoding guarantees. They also show that being
MDS(l) is equivalent to the generator matrix being able to achieve generi
c zero patterns. Using the GM-MDS theorem which says that any Reed-Solomon
codes achieve any generic zero pattern this implies that they generically
achieve list decoding capacity.\nGuo-Zhang proved punctured Reed-Solomon
codes can achieve list decoding capacity over quadratic in 'n' size fields
by introducing a slack to these notions (this was later strengthened to l
inear by an improved argument due to Alrabiah\, Guruswami\, and Li). This
talk will cover these exciting developments and recent works by the speake
r with Brakensiek\, Gopi\, and Zhang which develop this theory over codes
sampled from irreducible varieties\, prove a GM-MDS theorem for this setti
ng\, and prove that punctured AG codes are list decodable up to capacity o
ver constant size fields.\n \nShort Bio:\nManik Dhar is an Instructor (Po
stdoc) of Applied Mathematics at Massachusetts Institute of Technology. He
received his BTech at the Indian Institute of Technology - Bombay in 2016
\, MS at Stanford University in 2018\, and received his PhD from Princeton
in 2023 where he was advised by Zeev Dvir. He is broadly interested in di
screte mathematics and theoretical computer science.\n
URL:https://www.tcs.tifr.res.in/web/events/1419
DTSTART;TZID=Asia/Kolkata:20240618T160000
DTEND;TZID=Asia/Kolkata:20240618T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1435
DTSTAMP:20240619T094028Z
SUMMARY:Tree evaluation in space O(log n . log log n)
DESCRIPTION:Speaker: Bikshan Chatterjee (TIFR)\n\nAbstract: \nTree Evaluati
on was considered to be a candidate problem solvable in polynomial time bu
t not in log space. The natural algorithm for performing tree evaluation t
akes roughly log^2(n) space and was conjectured to be optimal. The belief
was based on an assumption that space being used for storing old values ca
nnot be used for new computation. Cook and Mertz showed this assumption to
be false\, in earlier work. In this paper\, they improve the space comple
xity to O(log n . log log n) using similar strategy of reusing space being
used for storing old values without erasing it.\n
URL:https://www.tcs.tifr.res.in/web/events/1435
DTSTART;TZID=Asia/Kolkata:20240621T140000
DTEND;TZID=Asia/Kolkata:20240621T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1436
DTSTAMP:20240622T142135Z
SUMMARY:Online Correlated Selection and related problems
DESCRIPTION:Speaker: Arghya Chakraborty (TIFR)\n\nAbstract: \nOnline Correl
ated Selection (OCS) is a novel online selection problem introduced by Mat
thew Fahrbach\, Zhiyi Huang\, Runzhou Tao\, Morteza Zadimoghaddam to obtai
n improved online algorithm for thee dge-weighted bipartite matching probl
em. In this talk\, I'll describe the OCS problem and describe an optimal a
lgorithm for the 2-way OCS.We will then discuss extensions of the OCS prob
lem as well as applications to the online edge-weighted bipartite matching
problem.\n
URL:https://www.tcs.tifr.res.in/web/events/1436
DTSTART;TZID=Asia/Kolkata:20240624T143000
DTEND;TZID=Asia/Kolkata:20240624T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1439
DTSTAMP:20240626T093506Z
SUMMARY:Private Circuits
DESCRIPTION:Speaker: Hari Krishnan P A (TIFR)\n\nAbstract: \nConsider an ad
versary which can look into a (boolean) circuit and runs an algorithm to m
ake inferences about the signals carried by the components inside the circ
uit. These algorithms are called side-channel attacks. In this talk\, we w
ill see a theoretical model for these attacks and a scheme to protect the
circuit against these attacks under the given model. This talk will be bas
ed on "Private circuits: Securing hardware against probing attacks" by Ish
ai\, Sahai and Wagner in CRYPTO 2003\, one of the first papers which intro
duced this problem.\n
URL:https://www.tcs.tifr.res.in/web/events/1439
DTSTART;TZID=Asia/Kolkata:20240627T160000
DTEND;TZID=Asia/Kolkata:20240627T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1440
DTSTAMP:20240627T094235Z
SUMMARY:On the Satisfaction Probability of k-CNF Formulas
DESCRIPTION:Speaker: Koduri Choudary (TIFR)\n\nAbstract: \nWe will be prese
nting a paper of Till Tantau (with the above title)\, that appeared in CCC
2022.\nThe Satisfaction Probability $\\sigma(\\phi)$ of a propositional f
ormula $\\phi$ is the probability that a random assignment of variables sa
tisfies the formula. This paper studies the complexity of the problem $k$S
AT-PROB$_{>\\delta}=\\{\\phi\\in k$CNF$|\\sigma(\\phi)>\\delta\\}$ for fix
ed $k\\in\\mathbb{N}$ and $\\delta\\in[0\,1]$. While the complexity of a f
ew examples were already known (eg: 3SAT-PROB$_{>0}\\in$ NP-Complete)\, Ak
mal and Williams have recently showed that 3SAT-PROB$_{>\\frac{1}{2}}\\in$
P and 4SAT-PROB$_{>\\frac{1}{2}}\\in$ NP-Complete. In this paper\, the au
thor gives a complete characterisation in the form of a trichotomy (i.e. $
k$SAT-PROB$_{>\\delta}$ is in AC$^0$ or NL-Complete or NP-Complete) based
on the parameters $k$ and $\\delta$.\n \n
URL:https://www.tcs.tifr.res.in/web/events/1440
DTSTART;TZID=Asia/Kolkata:20240628T140000
DTEND;TZID=Asia/Kolkata:20240628T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1438
DTSTAMP:20240626T045748Z
SUMMARY:Counting Problems in Graphical Models
DESCRIPTION:Speaker: Vidya Sagar Sharma (TIFR)\n\nAbstract: \nA graphical m
odel is a probabilistic model that uses a graph to represent conditional i
ndependence relations between random variables. We study acyclic directed
graphical models\, where a DAG is used to represent conditional independen
ce relations and causal relationships between random variables. In this mo
del\, a DAG $D(V = \\{v_1\, v_2\, \\ldots\, v_n\\}\, E)$ represents a prob
ability distribution $P$\, defined over a set of random variables $V$\, if
$P(v_1\, v_2\, \\ldots\, v_n) = \\prod_{i}{P(v_i | \\text{Parent}(v_i))}$
. A node $v_i$ is said to be a parent of $v_j$ in a DAG if $v_i \\rightarr
ow v_j$ is an edge in $G$. A DAG entails a conditional independence relati
on $X \\perp Y \\mid Z$ if all the probability distributions represented b
y the DAG satisfy the conditional independence relation. There can be more
than one DAG that entails the same set of conditional independence relati
ons. Such DAGs are said to be Markov equivalent. Markov equivalent DAGs be
long to the same equivalence class\, called a Markov equivalence class (ME
C)\, which is graphically represented by the union of the DAGs it contains
.Many interesting combinatorial problems related to MECs exist in the lite
rature. One such combinatorial problem is: Given the graphical representat
ion of an MEC\, find the size of the MEC. The problem arises from the fact
that a DAG is also used as a causal graph where a directed edge $X \\righ
tarrow Y$ implies that $X$ is a direct cause of $Y$\, and the size of an M
EC measures the uncertainty of the causal model when relying solely on obs
ervational data. In applications\, more information about the underlying D
AG than that encoded in the MEC may be available\, for example due to acce
ss to some special set of interventions\, or some domain-specific knowledg
e. Meek referred to this as \\emph{background knowledge}\, and modeled it
as a specification of the directions of some of the edges of the underlyin
g DAG. Wienöbst et al. show that counting such background knowledge-consi
stent DAGs of an MEC is \\#P-hard. In this talk\, we discuss an FPT algori
thm for this problem. Another interesting combinatorial problem is: Given
an undirected graph $G$\, count the number of MECs that have the same skel
eton as $G$. MECs with the same skeleton have statistical significance. In
this talk\, we discuss an FPT algorithm that counts MECs with the same sk
eleton. We also discuss an FPT algorithm that counts MECs with better time
complexity than the previous algorithm when the input graph is chordal. A
dditionally\, we discuss a polynomial algorithm to solve the problem when
the input graph is a tree.\n
URL:https://www.tcs.tifr.res.in/web/events/1438
DTSTART;TZID=Asia/Kolkata:20240628T160000
DTEND;TZID=Asia/Kolkata:20240628T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1431
DTSTAMP:20240627T051715Z
SUMMARY:Fair Sparse Regression with Clustering: An Invex Relaxation for a C
ombinatorial Problem
DESCRIPTION:Speaker: Adarsh Barik (National University of Singapore (NUS))\
n\nAbstract: \nIn this talk\, we discuss the problem of fair sparse regres
sion on a biased dataset where bias depends upon a hidden binary attribute
. The presence of a hidden attribute adds an extra layer of complexity to
the problem by combining sparse regression and clustering with unknown bin
ary labels. The corresponding optimization problem is combinatorial\, but
we propose a novel reformulation of it as an invex optimization problem. W
e show that the inclusion of the debiasing/fairness constraint in our mode
l has no adverse effect on the performance. Rather\, it enables the recove
ry of the hidden attribute. The support of our recovered regression parame
ter vector matches exactly with the true parameter vector. Moreover\, we s
imultaneously solve the clustering problem by recovering the exact value o
f the hidden attribute for each sample. Our method uses carefully construc
ted primal-dual witnesses to provide theoretical guarantees for the combin
atorial problem. We show that the sample complexity of our method is logar
ithmic in terms of the dimension of the regression parameter vector.\nShor
t Bio:\nAdarsh Barik is a postdoc at the Institute of Data Science at the
National University of Singapore. He did his PhD in Computer Science at Pu
rdue University and his undergrad at IIT Madras before that. His research
interests are broadly in theoretical and computational aspects of optimiza
tion\, machine learning\, information theory\, and high-dimensional data a
nalysis.\n
URL:https://www.tcs.tifr.res.in/web/events/1431
DTSTART;TZID=Asia/Kolkata:20240701T160000
DTEND;TZID=Asia/Kolkata:20240701T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1437
DTSTAMP:20240626T042022Z
SUMMARY:A Formal Approach to Exchange Design and Regulation
DESCRIPTION:Speaker: Suneel Sarswat (TIFR)\n\nAbstract: \n\n
URL:https://www.tcs.tifr.res.in/web/events/1437
DTSTART;TZID=Asia/Kolkata:20240705T113000
DTEND;TZID=Asia/Kolkata:20240705T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1442
DTSTAMP:20240704T065450Z
SUMMARY:(Δ + 1) Vertex Coloring in O(n) communication
DESCRIPTION:Speaker: Parth Mittal (University of Waterloo)\n\nAbstract: \nI
will talk about the communication complexity of (Δ+1) vertex coloring\,
where the edges of an n-vertex graph of maximum degree Δ are partitioned
between two players. I will show a randomized protocol which uses O(n) bit
s of communication and ends with both players knowing the coloring. Combin
ed with a folklore Ω(n) lower bound\, this settles the randomized communi
cation complexity of (Δ+1)-coloring up to constant factors.Based on joint
work with Maxime Flin.\n
URL:https://www.tcs.tifr.res.in/web/events/1442
DTSTART;TZID=Asia/Kolkata:20240705T143000
DTEND;TZID=Asia/Kolkata:20240705T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1443
DTSTAMP:20240709T062350Z
SUMMARY:Limitations of Massively Parallel Constant-time Algorithms
DESCRIPTION:Speaker: Vaibhav Krishan (Indian Institute of Technology\, Bomb
ay)\n\nAbstract: \nHow many processors are needed to compute a function in
constant-time by a massively parallel algorithm? We consider this questio
n with the lens of Boolean circuits. Here\, we compose logic gates\, acti
ng as processors\, with the number of layers corresponding to the running
time\, to compute a Boolean function.Hence\, the question we would like
to answer is: how many gates do we need in such a circuit to compute a ce
rtain function if the circuit only has a constant number of layers? In p
articular\, we are searching for explicit functions that require a very l
arge number of gates\, with a given set of choices for the logic gates.An
swers to such questions have implications in learning theory\, cryptograp
hy\, and the study of randomness. In this talk\, I will discuss past and
recent progress on such questions. Then\, I will talk about a recent high
er-order Fourier-analysis based approach\, proposed by Bhrushundi\, Hosse
ini\, Lovett and Rao (ITCS 2019)\, towards resolving a thirty-five years
old conjecture. Afterwards\, I will describe our contribution in developi
ng this approach\, where we propose studying the projection of an explici
t set of vectors towards resolving this conjecture. The talk will conclu
de with some implications towards other frontiers in the study of Boolean
circuits.The work I will discuss in this talk was in collaboration with P
rof. Sundar Vishwanathan at IIT Bombay.\nShort Bio:\nVaibhav Krishan is a
recent PhD graduate from IIT Bombay\, where he worked under the supervis
ion of Prof. Sundar Vishwanathan and Prof. Nutan Limaye. He has worked in
the industry as a quantitative trader\, and as a data scientist\, after
completing his undergrad. \n
URL:https://www.tcs.tifr.res.in/web/events/1443
DTSTART;TZID=Asia/Kolkata:20240709T160000
DTEND;TZID=Asia/Kolkata:20240709T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1449
DTSTAMP:20240710T110123Z
SUMMARY:Sperner's lemma and the equidissection of regular polygons
DESCRIPTION:Speaker: Pranshu Gaba (TIFR)\n\nAbstract: \nEquidissection of a
regular polygon is the problem of dissecting the polygon into triangles o
f equal area. Given a regular n-gon and a positive integer m\, we ask if i
t is possible to equidissect the n-gon into m triangles. To answer this qu
estion\, we make use of Sperner's lemma along with some elementary number
theory. Sperner's lemma is a major combinatorial result that has found app
lications in topics such as fixed-point computation and fair division.We s
hall see Sperner's lemma and its proof\, and use it to solve the equidisse
ction problem for regular polygons. We will also look at some polygons tha
t cannot be equidissected into m triangles for any positive m. Finally\, w
e will go over the equidissection problem for higher-dimensional hypercube
s.References:- Monsky\, P. (1970). On Dividing A Square Into Triangles. Th
e American Mathematical Monthly\, 77(2)\, 161–164. https://doi.org/10.1
080/00029890.1970.11992441- Stein\, S. (2004). Cutting a Polygon into Tria
ngles of Equal Areas. The Mathematical Intelligencer 26\, 17–21. https:
//doi.org/10.1007/BF02985395\n
URL:https://www.tcs.tifr.res.in/web/events/1449
DTSTART;TZID=Asia/Kolkata:20240712T143000
DTEND;TZID=Asia/Kolkata:20240712T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1434
DTSTAMP:20240613T064040Z
SUMMARY:Vigyan Vidushi 2024
DESCRIPTION:Speaker: \n\nAbstract: \nThe STCS Vigyan Vidushi 2024 program
me is a summer school to introduce women with undergraduate and postgradua
te background to advanced topics in theoretical computer and systems scien
ces. The two-week programme will be held during the period 15th July –
26th July 2024 at the TIFR Mumbai campus\, and will consist of short te
chnical courses taught by faculty from STCS\, TIFR. Lectures will be sup
plemented by problem-solving sessions\, to enhance students’ understandi
ng.\nhttps://www.tcs.tifr.res.in/~stcs-vv-24/\n \n
URL:https://www.tcs.tifr.res.in/web/events/1434
DTSTART;VALUE=DATE:20240715
DTEND;VALUE=DATE:20240727
LOCATION:TIFR\, Mumbai.
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1451
DTSTAMP:20240711T102852Z
SUMMARY:Matchings and Popularity
DESCRIPTION:Speaker: Kavitha Telikepalli (TIFR)\n\nAbstract: \nThe problem
of computing a stable matching in a bipartite graph is an old and well-st
udied problem. Gale and Shapley showed in 1962 that such a matching always
exists and can be efficiently computed. This is a classical result in alg
orithms with many applications in economics and computer science. Stabilit
y is a strong and rather restrictive notion. This talk will be on a relaxa
tion of stability called ‘popularity’ and we will see simple and effic
ient algorithms for some popular matching problems. No background in algor
ithms or matching theory will be assumed.\nShort Bio:\nKavitha is professo
r at the School of Technology and Computer Science at TIFR Mumbai and s
he is also the dean of our school. Prior to joining TIFR\, she was a facul
ty at the Indian Institute of Science (Bengaluru) and a postdoc at Max-Pla
nck Institute for Informatics\, Saarbrücken. Her primary interests are in
the graph algorithms and combinatorial optimisation\, and has made founda
tional contributions in the area of graph matchings.\n
URL:https://www.tcs.tifr.res.in/web/events/1451
DTSTART;TZID=Asia/Kolkata:20240715T160000
DTEND;TZID=Asia/Kolkata:20240715T170000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1447
DTSTAMP:20240710T065515Z
SUMMARY:Verification of Concurrent Programs Under Weak Memory
DESCRIPTION:Speaker: Krishna S. (Indian Institute of Technology\, Bombay)\n
\nAbstract: \nThis is an overview of recent work on the verification of co
ncurrent programs. Traditionally concurrent programs are interpreted under
sequential consistency (SC). Even though SC is very intuitive and easy to
use\, modern multiprocessors do not employ SC for performance reasons\, a
nd instead use so-called ‘weak memory models’. Some of the well known
weak memory models in vogue among modern multiprocessor architectures are
Intel x-86\, IBM POWER and ARM. The use of weak memory is also prevalent i
n the C11 model\, leading to the release acquire fragment of C11. In this
talk\, I will introduce some of these memory models\, as well as recent ef
forts in the verification of programs under these.\nShort Bio:\nKrishna is
a professor at the Department of Computer Science and Engineering at th
e Indian Institute of Technology\, Bombay. She obtained her PhD from IIT
Madras. Her primary interests are in the areas of automata theory\, logic\
, and formal verification of timed and probabilistic systems.\n
URL:https://www.tcs.tifr.res.in/web/events/1447
DTSTART;TZID=Asia/Kolkata:20240718T160000
DTEND;TZID=Asia/Kolkata:20240718T170000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1453
DTSTAMP:20240719T064159Z
SUMMARY:Catch them if you can
DESCRIPTION:Speaker: Juhi Chaudhary (TIFR)\n\nAbstract: \nThe Cops and Robb
er game is a well-studied two-player pursuit-evasion game played on graphs
\, where a team of cops attempts to capture a robber. The cop number of a
graph represents the minimum number of cops required for a successful capt
ure. Graphs with a cop number of one are known as cop-win graphs. In this
seminar\, I will begin by characterizing cop-win graphs. Subsequently\, I
will demonstrate\, using the technique of guarding a subgraph—a method f
or bounding the cop number of graphs with geometric representations—that
the cop number for planar graphs is atmost three.\n
URL:https://www.tcs.tifr.res.in/web/events/1453
DTSTART;TZID=Asia/Kolkata:20240720T143000
DTEND;TZID=Asia/Kolkata:20240720T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1445
DTSTAMP:20240725T063708Z
SUMMARY:Using AI to assist in improving maternal and child health outcomes
in underserved communities in India
DESCRIPTION:Speaker: Aparna Taneja (Google Research India)\n\nAbstract: \n\
nThe widespread availability of cell phones has enabled non-profits to del
iver critical health information to their beneficiaries in a timely manner
. This project assists non-profits that employ automated messaging program
s to deliver timely preventive care information to beneficiaries (new and
expecting mothers) during pregnancy and after delivery. Unfortunately\, a
key challenge in such information delivery programs is that a significant
fraction of beneficiaries drop out of the program. Yet\, non-profits often
have limited health-worker resources (time) to place crucial service call
s for live interaction with beneficiaries to prevent such engagement drops
. To assist non-profits in optimizing this limited resource\, we developed
a Restless Multi-Armed Bandits (RMABs) system. The RMAB system was evalua
ted in collaboration with an NGO via a real-world service quality improvem
ent study and showed a 30% reduction in engagement drops. This model was e
ventually deployed by the NGO and has served over 350K women so far. These
encouraging results have led to a new collaboration with the Kilkari prog
ram\, the largest maternal mHealth program in the world. And we hope to se
e similar improvements in engagement with the use of our AI model.\n \nS
hort Bio:\nAparna is a researcher at the Multi Agents Systems for Social I
mpact team in Google Research India. She received her PhD in Computer Sci
ence at ETH Zurich under the supervision of Prof. Marc Pollefeys. She then
pursued a postdoc at Disney Research Zurich. She collaborates with severa
l NGO’s and academic partners in the fields of public health and conserv
ation and her primary focus is collaboration with ARMMAN\, an NGO focused
on improving maternal and child health outcomes in underserved communitie
s in India.\n
URL:https://www.tcs.tifr.res.in/web/events/1445
DTSTART;TZID=Asia/Kolkata:20240725T170000
DTEND;TZID=Asia/Kolkata:20240725T180000
LOCATION:AG-66
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1448
DTSTAMP:20240726T041235Z
SUMMARY:Faul Tolerant Distance Oracles
DESCRIPTION:Speaker: Dipan Dey (IIT Gandhinagar)\n\nAbstract: \nThe shortes
t distance and shortest paths between vertices are very important aspects
of Graph Theory. Graphs are often used to represent real-life networks and
real-life networks are often prone to failures. Due to those failures\, w
e may want to avoid some edges in the graph at some point of time. \nHenc
e\, we may want to find the shortest distance and the shortest path betwee
n two vertices while avoiding some edges or vertices in the graph. Fault-t
olerant distance oracles are oracles or a set of data structures which can
answer those kinds of queries. In my talk\, I will be discussing some of
our results related to fault-tolerant distance oracles.\nShort Bio: \nDipa
n obtained a B.Sc. (Hons.) in Mathematics from Vidyasagar College (Kolkata
) and an M.Sc. in Mathematics from Banaras Hindu University. During his po
stgraduate studies at Banaras Hindu University\, he developed an interest
in graph theory. He then joined the CSE discipline at IIT Gandhinagar to p
ursue a PhD under Prof. Manoj Gupta's supervision. His thesis focuses on f
ault-tolerant distance oracles\, the data structures that can determine di
stances between vertices while avoiding some edges or vertices. In additio
n to his thesis work\, he has also worked on linear rank width one graphs
and power graphs.\n \n
URL:https://www.tcs.tifr.res.in/web/events/1448
DTSTART;TZID=Asia/Kolkata:20240726T100000
DTEND;TZID=Asia/Kolkata:20240726T110000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1455
DTSTAMP:20240722T064620Z
SUMMARY:Bandit Problems with Rare Rewards
DESCRIPTION:Speaker: Anirban Bhattacharjee (TIFR)\n\nAbstract: \nIn this ta
lk\, we will examine bandit algorithms in the special and relevant case of
rarely occuring rewards. This problem will be seen from the perspective o
f both fixed confidence and fixed budget settings. In both these settings\
, we have been able to use approximate methods to drastically reduce the c
omputational complexity of existing algorithms. We have also devised an al
gorithm to select the best system from a given collection of highly reliab
le systems\, where failures are rare events. \n
URL:https://www.tcs.tifr.res.in/web/events/1455
DTSTART;TZID=Asia/Kolkata:20240726T110000
DTEND;TZID=Asia/Kolkata:20240726T120000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1457
DTSTAMP:20240724T062538Z
SUMMARY:A randomized algorithm for estimate volumes of convex
DESCRIPTION:Speaker: Sourav Roy (TIFR)\n\nAbstract: \nI will present a ran
domized volume algorithm from the paper "Volume Estimates and Rapid Mixi
ng " by Bela Bollobas. The goal is to find a fully polynomial approximatio
n scheme (FPRAS) for approximating the volume of a convex body K in R^n. A
randomized algorithm that runs in time polynomial in \, 1/ (\\epsilon)
and log(1/\\nu)\, and with probability at least (1 − \\nu) produces a
n \\epsilon-approximation to vol(K). Here the input is (r\, R\, n) for K
satisfying rB^n \\subset K \\subset RB^n with = n +
+ \, where is the number of binary digits of a dyadic rational x.\n
URL:https://www.tcs.tifr.res.in/web/events/1457
DTSTART;TZID=Asia/Kolkata:20240726T160000
DTEND;TZID=Asia/Kolkata:20240726T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1458
DTSTAMP:20240726T100756Z
SUMMARY:Optimal Strategies and Minimax Lower Bounds for Online Convex Games
DESCRIPTION:Speaker: Pranab Panda (TIFR)\n\nAbstract: \nhttps://www.learnin
gtheory.org/colt2008/papers/111-Abernethy.pdf\n
URL:https://www.tcs.tifr.res.in/web/events/1458
DTSTART;TZID=Asia/Kolkata:20240729T160000
DTEND;TZID=Asia/Kolkata:20240729T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1430
DTSTAMP:20240722T043818Z
SUMMARY:Type-Checking for Pattern-Based Tree Transformations
DESCRIPTION:Speaker: M. Praveen (Chennai Mathematical Institute)\n\nAbstrac
t: \nWe introduce and study pattern-based tree transformations. Consider (
+(.(x\, y)\, .(x\, z))\, .(x\, +(y\, z)))\, a pair of a source pattern and
a target pattern. This source pattern matches any expression e of the for
m +(.(e1 \, e2 )\, .(e1 \, e3 ))\, by substituting e1 for x\, e2 for y and
e3 for z. Note that the set of such matching expressions is not a regular
tree language\, since the same expression must appear in two locations wh
ere x appears in the source pattern. This pair transforms the expression e
into the expression .(e1 \, +(e2 \, e3 )) (obtained by applying the same
substitution to the target pattern). Our model finitely represents a (poss
ibly infinite) set of such (source pattern\, target pattern) pairs. The ex
pressive power of this model comes at the cost of undecidability of checki
ng equivalence. The type-checking problem checks whether applying a given
transformation to trees having a given regular property (type) preserves t
he property. We show that this problem is decidable for pattern-based tree
transformations. Our decision procedure is by a reduction to the emptines
s problem of alternating tree automata.This is joint work with C. Aiswarya
and Sahil Mhaskar from Chennai Mathematical Institute.\nShort Bio:\nPrave
en is an associate professor at Chennai Mathematical Institute. He finishe
d PhD at the Institute of Mathematical Institute\, Chennai. He was a postd
oc at LSV\, ENS Cachan and LaBRI\, University of Bordeaux before he joined
CMI in 2014. His research interests and logic and automata theory for for
mal verification.\n
URL:https://www.tcs.tifr.res.in/web/events/1430
DTSTART;TZID=Asia/Kolkata:20240730T160000
DTEND;TZID=Asia/Kolkata:20240730T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1459
DTSTAMP:20240801T054230Z
SUMMARY:EFX Allocations on Graphs.
DESCRIPTION:Speaker: Yeshwant Chandrakant Pandit (TIFR)\n\nAbstract: \nIn
this talk\, we will explore envy-freeness up to any good (EFX) in settings
where valuations are represented by a graph of arbitrary size\, with vert
ices corresponding to agents and edges corresponding to items. An item (ed
ge) has zero marginal value to all agents (vertices) not incident to the e
dge. Each vertex may have an arbitrary monotone valuation on the set of in
cident edges. We will first examine allocations that correspond to edge or
ientations\, noting that EFX may not always be achievable. Furthermore\, d
etermining whether an EFX orientation exists is NP-complete. The main resu
lt is that EFX allocations exist for this setting. This is one of the few
cases where EFX allocations are known to exist for more than 3 agents.\nRe
ference: https://dl.acm.org/doi/pdf/10.1145/3580507.3597764\n \n
URL:https://www.tcs.tifr.res.in/web/events/1459
DTSTART;TZID=Asia/Kolkata:20240802T143000
DTEND;TZID=Asia/Kolkata:20240802T153000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1441
DTSTAMP:20240722T043611Z
SUMMARY:Equivalence Testing of Principal Minors
DESCRIPTION:Speaker: Roshan Raj (IIT Bombay)\n\nAbstract: \nFor a square ma
trix A with n rows and n columns and a subset S of [n]\, the corresponding
principal minor is the determinant of the submatrix of A with rows and co
lumns indexed by elements in S i.e. det(A[S\, S]). Two operations on a mat
rix A that preserve all the principal minors are:\nTaking the transpose of
the matrix.\nMultiplying it by an invertible diagonal matrix on one side
and by its inverse on the other.\nThe matrices DAD^{-1} and DA^TD^{-1} are
called diagonally similar to A when D is an invertible diagonal matrix.\n
For a square matrix A of dimension n\, a subset S of [n] of size at least
two and at most n-2 is called a cut if the rank of both the submatrices A[
S\, [n]-S] and A[[n]-S\, S] is less than two. In a couple of works\, Rapha
el and Loewy showed that for an irreducible matrix A with no cuts\, anothe
r matrix B has the same principal minors if and only if B is diagonally si
milar to A. \nIn this work\, we give an extension of their result by givi
ng a complete characterization of when two matrices can have the same prin
cipal minors and give a polynomial time algorithm to test it. We also pres
ent some applications of equivalence testing of principal minors in Combin
atorics and Polynomial Identity Testing.\nThis is joint work with Abhranil
Chatterjee\, Rohit Gurjar\, and Sumanta Ghosh.\nShort Bio:\nRoshan Raj
is a PhD student in the CSE department at IIT Bombay working under the me
ntorship of Prof. Rohit Gurjar. He holds a Betch degree in CSE from IIT B
HU.\n
URL:https://www.tcs.tifr.res.in/web/events/1441
DTSTART;TZID=Asia/Kolkata:20240802T160000
DTEND;TZID=Asia/Kolkata:20240802T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1424
DTSTAMP:20240801T044101Z
SUMMARY:Invitation to FPT Approximation via Cut and Coverage Problems
DESCRIPTION:Speaker: Saket Saurabh (The Institute of Mathematical Sciences)
\n\nAbstract: \nIn this talk\, we will provide an introduction to the area
of FPT-Approximation. The goal of this research area is to develop algori
thms that are faster than known FPT algorithms and achieve better approxim
ation factors than what is possible in polynomial time. We will explore re
cent advances in this field through some classical cut and coverage proble
ms\, such as k-Way Cut and Max-Coverage.\nShort Bio:\nSaket Saurabh receiv
ed his PhD in Computer Science (2008)\, from The Institute of Mathe-matic
al Sciences (IMSc)\, Chennai. Saurabh spent two years (2007-2009) as a Po
stdoctoral Fellow at University of Bergen\, Norway\, and is now a profess
or at IMSc. He is a SwarnaJayanti Fellow in Mathematical Sciences (2018)
\, Fellow of Indian Academy of Sciences (2020)\, Academia Europaea (2020)\
, and European Association for Theoretical Computer Science (EATCS\, 2021
). He received the inaugural ACM India Early Career Researcher Award in 2
020\, and Shanti Swarup Bhatnagar Prize (SSB) for Science and Technology
2021 (Mathematical Science).\n
URL:https://www.tcs.tifr.res.in/web/events/1424
DTSTART;TZID=Asia/Kolkata:20240806T160000
DTEND;TZID=Asia/Kolkata:20240806T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1452
DTSTAMP:20240801T044137Z
SUMMARY:Advances and Challenges in Fair Division: Quantiles\, Subsidies\, a
nd Randomization
DESCRIPTION:Speaker: Vishnu V Narayan (McGill University\, Canada)\n\nAbstr
act: \nThe question of how to fairly divide a collection of indivisible it
ems amongst a set of agents has remained of central importance to humanity
since antiquity. In this fundamental problem\, the agents have varied pre
ferences\, and an allocator seeks to find a single allocation such that ev
ery agent perceives its bundle as fair. This problem arises in various app
lications\, ranging from classical examples like the division of inherited
estates and international borders to modern applications such as assignin
g seats in college courses and allocating computational resources fairly.R
ecent decades have witnessed significant progress\, transforming this prob
lem into a fascinating mathematical landscape with surprising results and
intriguing new challenges. The broad goal of the community is to devise de
finitions of fairness that mirror our intuitive understanding of what it m
eans to be fair\, and then study questions such as: does a fair allocation
always exist?\; can one be (efficiently) computed?\; what are the precise
limits to the degree of fairness one can guarantee? Fair division is emer
ging as a major research area\, with an increasing number of publications
at Theory and AI conferences each year. In this talk\, I will focus on sel
ected recent papers of mine\, highlighting three techniques (Quantiles\, S
ubsidies\, and Randomization) we use to extend the study of fair allocatio
ns to general valuation classes and resolve some conjectures and open prob
lems. This talk will also provide an overview of my research trajectory an
d plans for future work.\nShort Bio:\nVishnu V. Narayan is a postdoctoral
fellow hosted by Michal Feldman at Tel Aviv University. His main research
focus is on the fair division of indivisible items. He is also broadly int
erested in the many intersections of combinatorics\, algorithms\, and game
theory. He has a best paper award from SAGT 2019 and a Highlights Session
selection at EC 2024. Earlier\, he completed his Ph.D. in Computer Scienc
e at McGill University under the supervision of Adrian Vetta\, and his M.M
ath. in Combinatorics and Optimization at the University of Waterloo with
Joseph Cheriyan.\n
URL:https://www.tcs.tifr.res.in/web/events/1452
DTSTART;TZID=Asia/Kolkata:20240807T110000
DTEND;TZID=Asia/Kolkata:20240807T120000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1461
DTSTAMP:20240808T091729Z
SUMMARY:Regret minimization in stochastic multi-armed bandits
DESCRIPTION:Speaker: Agniv Bandyopadhyay (TIFR)\n\nAbstract: \nIn the stoch
astic K-armed bandit framework\, we are given K unknown distributions or a
rms. At a given time\, we can select one arm and can observe one sample fr
om that arm. Our goal is to maximize the reward over a finite time horizon
\, which is also equivalent to minimizing the regret. Regret minimization
is an important aspect of many applications where we have to make sequenti
al decisions under uncertainty and optimize for some objective\, for examp
le\, clinical trials\, recommendation systems\, selecting the best portf
olio in a financial market\, etc. We will analyze some basic regret minimi
zing algorithms\, such as: explore-then-commit\, successive reject\, upper
confidence bound\, etc. We will also derive an information-theoretic lowe
r bound on regret. \n \nThe talk will be based on the following referenc
es: \n1. Lattimore\, Tor\, and Csaba Szepesvári. Bandit algorithms. Camb
ridge University Press\, 2020\, Chapter 6-7.\n2. Kaufmann\, Emilie. Contri
butions to the Optimal Solution of Several Bandit Problems. Diss. Universi
té de Lille\, 2020\, Chapter 1.\n
URL:https://www.tcs.tifr.res.in/web/events/1461
DTSTART;TZID=Asia/Kolkata:20240809T160000
DTEND;TZID=Asia/Kolkata:20240809T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1460
DTSTAMP:20240807T043612Z
SUMMARY:STCS Annual Symposium
DESCRIPTION:Speaker: \n\nAbstract: \nThe STCS Symposium\, to be held on Aug
ust 12-13 this year\, is an annual event where members showcase their rese
arch work. This year's edition has several invited talks including the pre
stigious Prof. R. Narasimhan Memorial Lecture.\nVisit STCS Annual Symposiu
m 2024 page for more details.\n
URL:https://www.tcs.tifr.res.in/web/events/1460
DTSTART;VALUE=DATE:20240812
DTEND;VALUE=DATE:20240814
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1446
DTSTAMP:20240812T062023Z
SUMMARY:Recent progress on the online k-server problem
DESCRIPTION:Speaker: Amit Kumar (IIT Delhi)\n\nAbstract: \nThe k-server pro
blem is one of the most fundamental problems in the area of online algorit
hms. In this problem\, an online algorithm needs to maintain a set of k se
rvers in a metric space. When a request arrives at a certain location\, on
e of the k servers needs to move to this requested location. The goal is t
o minimize the total movement cost of all the servers. Over the past three
decades\, significant progress has been made on this problem\, yet many i
ntriguing questions remain open. Notably\, progress on the k-server proble
m has resulted in the development of new techniques that have advanced the
broader field of online algorithms. This talk will survey recent techniqu
es and breakthroughs developed for the k-server problem.\nShort Bio:\nAmit
Kumar is a faculty member in the dept. of Computer Science and Engineeri
ng at IIT Delhi. He obtained a B.Tech. degree from IIT Kanpur in 1997 and
Ph.D. from Cornell University in 2002. He works in the area of combinato
rial optimization\, with emphasis on problems arising in scheduling\, gra
ph theory and clustering. Hereceived IBM Faculty Award in 2005\, INAE (Ind
ian National Academy of Engineering) Young Engineer Award in 2006 and INS
A (Indian National Science Academy) Medal for Young Scientists in 2011. H
e was a Max Planck-India partner group research fellow during 2005-09. He
received the prestigious Shanti Swarup Bhatnagar Award for Mathematical
Sciences in 2018\, and was elected Fellow of Indian Academy of Sciences i
n 2019\, and Fellow of Indian National Academy of Engineering in 2022.\n
URL:https://www.tcs.tifr.res.in/web/events/1446
DTSTART;TZID=Asia/Kolkata:20240812T090000
DTEND;TZID=Asia/Kolkata:20240812T103000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1418
DTSTAMP:20240812T062049Z
SUMMARY:Verifying Programs in Weak Memory Models with persistency
DESCRIPTION:Speaker: Prakash Saivasan (The Institute of Mathematical Scienc
es)\n\nAbstract: \nIn this talk\, we will consider the problem of verifyin
g concurrent programs. In here\, we are given a set of programs that commu
nicate through shared memory\, a specification and we wish algorithmically
check if the programs violate the specification. The programmers\, while
writing code usually assume that the memory operations are immediate (refe
rred to as sequential consistency). However\, the modern day architectures
\, to optimise the running time\, re-order the memory operations in a non-
trivial manner. This leads to various memory models such as TSO\, PSO and
so on. We will walk through some of these memory models during the talk. T
he recent intel processor introduced persistency mechanism that allows for
the writes to be archived. This can then be used to restart the computati
on in case of a crash. The main focus of the talk will be how to verify pr
ograms when persistency is combined with weak memory.\nShort Bio:\nPrakash
Saivasan | The Institute of Mathematical Sciences (imsc.res.in)\n
URL:https://www.tcs.tifr.res.in/web/events/1418
DTSTART;TZID=Asia/Kolkata:20240813T093000
DTEND;TZID=Asia/Kolkata:20240813T103000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1456
DTSTAMP:20240812T062115Z
SUMMARY:The Compensated Coupling (or How the Future is the Great Guide for
the Present)
DESCRIPTION:Speaker: Siddhartha Banerjee (Cornell University)\n\nAbstract:
\nI will present the compensated coupling: a simple paradigm for designing
sequential decision-making policies based on sample-pathwise comparisons
against a hindsight benchmark. This approach generalizes many standard res
ults used in studying Markov decision processes and reinforcement learning
\, but also gives us new policies which are much simpler and more effectiv
e than existing heuristics. For a large class of widely-studied sequential
decision-making problems -- including network revenue management\, dynami
c pricing\, generalized assignment\, online bin packing\, online assortmen
t optimization and bandits with knapsacks -- I will illustrate how under a
wide range of conditions\, our approach achieves additive loss compared t
o the hindsight optimal which is independent of the horizon and state-spac
e. Time permitting\, I will try and describe how we can use this technique
to incorporate side information and historical data\, and achieve constan
t regret with as little as a single data trace.\nShort Bio:\nSid Banerjee
is an associate professor in the School of Operations Research at Cornell\
, working on topics at the intersection of data-driven decision-making\, n
etwork algorithms and market design. His research is supported by grants f
rom the NSF (including an NSF CAREER award)\, ARO\, AFOSR\, and Engaged Co
rnell\, and has received multiple awards including the INFORMS Applied Pro
bability Society Best Publication award in 2021 and the Erlang Prize in 20
22. He received his B.Tech from IIT Madras and PhD from the ECE Department
at UT Austin\, and was a postdoctoral researcher in the Social Algorithms
Lab at Stanford. He also served as a technical consultant with the resear
ch science group at Lyft from 2014-18.\n
URL:https://www.tcs.tifr.res.in/web/events/1456
DTSTART;TZID=Asia/Kolkata:20240813T143000
DTEND;TZID=Asia/Kolkata:20240813T153000
LOCATION:AG-69
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1463
DTSTAMP:20240816T063415Z
SUMMARY:EFX Allocation Exists for Three Agents
DESCRIPTION:Speaker: Soumyajit Pyne (TIFR)\n\nAbstract: \n In 2020\, Chaud
hury\, Garg\, and Mehlhorn demonstrated the existence of EFX allocation fo
r three agents with additive valuations\, though their proof is complex an
d lengthy. Subsequently\, Akrami et al. developed a simpler proof of this
result. In the seminar talk\, we will review their proof. As of now\, the
existence of EFX for more than three agents remains unknown.\nLink to the
paper: https://arxiv.org/abs/2205.07638.\n \n
URL:https://www.tcs.tifr.res.in/web/events/1463
DTSTART;TZID=Asia/Kolkata:20240816T160000
DTEND;TZID=Asia/Kolkata:20240816T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1444
DTSTAMP:20240812T061508Z
SUMMARY:Exponential lower bounds via Exponential sums
DESCRIPTION:Speaker: Somnath Bhattacharjee (Chennai Mathematical Institute)
\n\nAbstract: \nValiant's famous VP vs. VNP conjecture states that the sym
bolic permanent polynomial does not have polynomial-size algebraic circuit
s. However\, the best upper bound on the size of the circuits computing th
e permanent is exponential. Informally\, VNP is an exponential sum of VP-c
ircuits. In this paper we study whether\, in general\, exponential sums of
algebraic circuits require exponential-size algebraic circuits. We show t
hat the famous Shub-Smale tau-conjecture indeed implies such an exponentia
l lower bound for an exponential sum. Our main tools come from parameteriz
ed complexity. Along the way\, we also prove an exponential fpt (fixed-par
ameter tractable) lower bound for the parameterized algebraic complexity c
lass VW[P]\, weighted sum circuits (constant-free\, unbounded degree)\, as
suming the same conjecture. VW[P] can be thought of as the weighted sums o
f (unbounded-degree) circuits\, where only +1/-1 constants are cost-free.
To the best of our knowledge\, this is the first time the Shub-Smale tau-c
onjecture has been applied to prove explicit exponential lower bounds. Fur
thermore\, we prove that when this class is fpt\, then a variant of the co
unting hierarchy\, namely the linear counting hierarchy collapses. Moreove
r\, if a certain type of parameterized exponential sums is fpt\, then inte
gers\, as well as polynomials with coefficients being definable in the lin
ear counting hierarchy have subpolynomial tau-complexity. Finally\, we sho
wed a completeness result on a subclass VW[F] (weighted sum of formulas)\n
\nThis talk is based on joint work with Markus Blaser (University of Saa
rland)\, Pranjal Dutta (NUS) and Saswata Mukherjee (NUS) [ICALP 2024]\, li
nk: https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ICALP.2024
.24\n \nShort Bio:\nSomnath completed his Bachelors and Masters from the
Chennai Mathematical Institute and is on his way to the University of Toro
nto for his PhD. His current research interests are in algorithms and comp
lexity with a focus on problems with an algebraic flavor\, including algeb
raic complexity\, algebraic algorithms\, pseudorandomness\, algebraic proo
f complexity and error correcting codes.\n
URL:https://www.tcs.tifr.res.in/web/events/1444
DTSTART;TZID=Asia/Kolkata:20240819T160000
DTEND;TZID=Asia/Kolkata:20240819T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1433
DTSTAMP:20240819T052756Z
SUMMARY:Wait ... even my phone charger can steal my data?
DESCRIPTION:Speaker: Purushottam Kar (IIT Kanpur)\n\nAbstract: \nIn the con
text of cybersecurity\, a side-channel attack operates by targeting the im
plementation of an algorithm or protocol rather than identifying flaws in
the algorithm or protocol itself. We explore a family of such attacks\, ut
ilizing the electromagnetic (EM) or power channels to perform exfiltration
i.e.\, steal data from a device. Specifically\, we study the exfiltration
of AI/ML models operating inside a device using a limited number of "prob
es". Our path to this nefarious goal takes us through the seemingly unrela
ted problem of how permutations induced by the quotient groups Z/NZ and (Z
/NZ)^x act upon the Hamming topology. We arrive at novel results in this d
irection that allow disturbingly efficient attacks using a small number of
probes. These attacks target simple arithmetic operations such as additio
n or multiplication carried out while an AI/ML model is performing inferen
ce. The talk will present a short guided tour of this study and conclude w
ith some fascinating yet open questions thereby surfaced.\nShort Bio:\nRoo
p is a member of the IIT Kanpur faculty. He finds the IITK campus so nice
that he graduated from it twice. He is fascinated by machine learning and
optimization and has thus far managed to convince his employers that he un
derstands these areas decently. Consequently\, they let him stick around w
hile he explores ways to use ML to achieve clean-air objectives\, assist i
n teaching comically large class sizes\, reveal and mitigate cybersecurity
threats\, and make rich corporations even richer.\n
URL:https://www.tcs.tifr.res.in/web/events/1433
DTSTART;TZID=Asia/Kolkata:20240820T160000
DTEND;TZID=Asia/Kolkata:20240820T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1465
DTSTAMP:20240821T100350Z
SUMMARY:Making a Trifference
DESCRIPTION:Speaker: Siddharth Bhandari (Toyota Technological Institute)\n\
nAbstract: \n\nA subset C⊆{0\,1\,2}^n is said to be a trifferent code (o
f block length n) if for every three distinct codewords x\,y\,z∈C\, ther
e is a coordinate i∈{1\,2\,...\,n} where they all differ\, that is\, {x(
i)\,y(i)\,z(i)} is same as {0\,1\,2}. Let T(n) denote the size of the larg
est trifferent code of block length n. Understanding the asymptotic behavi
or of T(n) is closely related to determining the zero-error capacity of th
e (3/2)-channel defined by Elias'58\, and is a long-standing open problem
in the area. Elias had shown that T(n)≤2×(3/2)^n and prior to our work
the best upper bound was T(n)≤0.6937×(3/2)^n due to Kurz'23 obtained vi
a a computer search. \n \nIn this talk we will see an improved bound of
T(n)≤c×n^(−0.4)×(3/2)^n where c is an absolute constant. First\, we
will go over some history of the problem and explore its connections to ze
ro-error information theory\, perfect hashing and graph covering. Then\, w
e will go over the main ideas of the proof and several interesting open qu
estions.\nShort Bio:\nSiddharth Bhandari is currently serving as a Researc
h Assistant Professor at the Toyota Technological Institute in Chicago. Pr
ior to this\, he was a Simons-Berkeley Fellow at the Simons Institute for
the Theory of Computing. Siddharth completed his Ph.D. at the School of Te
chnology and Computer Science at the Tata Institute of Fundamental Researc
h\, Mumbai. His research interests lie in exploring interdisciplinary dire
ctions with a focus on developing techniques from computer science. He has
worked and continues to work in the areas of coding/information theory\,
MCMC sampling algorithms\, and causal inference. Siddharth's work has b
een recognized with several awards. His dissertation work was awarded the
ACM India Dissertation Award. His paper "Improved Bounds for Perfect Sampl
ing of k-Colorings in Graphs" won the Danny Lewin Best Student Paper Award
at STOC 2020. He also won the Jack Keil Wolf Student Paper Award at ISIT
2018 for his work "Bounds on the Zero-Error List-Decoding Capacity of the
q/(q−1) Channel.\n
URL:https://www.tcs.tifr.res.in/web/events/1465
DTSTART;TZID=Asia/Kolkata:20240826T100000
DTEND;TZID=Asia/Kolkata:20240826T110000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1417
DTSTAMP:20240827T094728Z
SUMMARY:Colorful Extensions of Infinite (p\,q)-Theorems in Combinatorial Ge
ometry
DESCRIPTION:Speaker: Arijit Ghosh (Indian Statistical Institute)\n\nAbstrac
t: \nAn infinite sequence of sets is said to be a heterochromatic sequ
ence for an infinite collection of families of sets\, if there exists
a strictly increasing sequence of natural numbers such that for all
we have . In this talk\, we will prove that if for each is a family
of nicely shaped convex sets in such that each heterochromatic sequen
ce of contains sets that can be pierced by a single -flat ( -d
imensional affine space) then all but finitely many 's can be pierced by
finitely many -flats. This result generalizes the -Theorem proved by
Keller and Perles (SoCG'22) to the countably colorful setting. We have a
lso established the tightness of our results by proving several no-go the
orems.\nThis is a joint work with Sutanoya Chakraborty (PhD Student at ISI
\, Kolkata) and Soumi Nandi (PhD Student at ISI\, Kolkata).\nShort Bio:\nA
rijit Ghosh is currently an Associate Professor at ACM Unit\, Indian Stati
stical Institute\, Kolkata. He did his PhD in Computer Science from INRIA\
, France\, and was a Postdoc at Max Planck Insitute for Informatics\, Ger
many.\n \n
URL:https://www.tcs.tifr.res.in/web/events/1417
DTSTART;TZID=Asia/Kolkata:20240827T160000
DTEND;TZID=Asia/Kolkata:20240827T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1470
DTSTAMP:20240828T041544Z
SUMMARY:An introduction to fairness in reinforcement learning.
DESCRIPTION:Speaker: Vishakha Patil (Indian Institute of Science\, Bangalor
e)\n\nAbstract: \nIn a series of three talks\, we present an overview of w
ork in bandit optimization\, including stochastic and adversarial bandits\
, and will focus on variants of multi-armed bandits where fairness is an a
dditional constraint. In Fair-MAB\, for example\, in addition to the objec
tive of maximizing the sum of expected rewards\, the algorithm also needs
to ensure that at any time\, each arm is pulled at least a pre-specified f
raction of times. We investigate the interplay between learning and fairne
ss in such settings\, and also investigate the cost of fairness.Talk 1: Th
ursday 29/8\, 2 - 3:30 pm in A-238Talk 2: Friday 30/8\, 4 - 5:30 pm in A-2
01Talk 3: Monday 2/9\, 2 - 3:30 pm in A-238\nShort Bio:\nVishakha Patil wo
rks in online learning and optimization\, game theory\, and mechanism desi
gn. She completed her PhD from IISc\, supported by a Google Fellowship and
a CII-SERB PMRF. She has also been a Heidelberg Laureate Forum Young Rese
archer\, and her MTech thesis received an Honourable Mention for the Best
Thesis Award at CSA\, IISc.\n
URL:https://www.tcs.tifr.res.in/web/events/1470
DTSTART;TZID=Asia/Kolkata:20240829T140000
DTEND;TZID=Asia/Kolkata:20240829T150000
LOCATION:A-238
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1472
DTSTAMP:20240830T040206Z
SUMMARY:Adversarial Multi-Armed Bandit and the EXP3 Algorithm.
DESCRIPTION:Speaker: Vishakha Patil (Indian Institute of Science\, Bangalor
e)\n\nAbstract: \nIn this talk\, we will look at the Adversarial Multi-Arm
ed Bandit problem. In this model\, as the name suggests\, the rewards are
chosen by an adversary. We then present the EXP3 algorithm\, a well-known
algorithm for regret minimization in Adversarial Bandits\, and analyze its
regret.\n
URL:https://www.tcs.tifr.res.in/web/events/1472
DTSTART;TZID=Asia/Kolkata:20240830T160000
DTEND;TZID=Asia/Kolkata:20240830T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1471
DTSTAMP:20240830T053500Z
SUMMARY:Public Quantum Network: The First Node
DESCRIPTION:Speaker: Keshav Kapoor (University of Illinois Urbana-Champaign
)\n\nAbstract: \nQuantum networks are being developed around the globe in
order to facilitate cryptographic systems\, improve metrology\, and advanc
e quantum computing systems. These networks have yet to be made readily av
ailable to the general public. We have developed a publicly accessible qua
ntum network node\, located in Urbana-Champaign. We utilize a polarizati
on-based entanglement source located at the University of Illinois Urbana-
Champaign. One of the photons from the source is routed through fiber opti
c cables to the Urbana Free Library\, a public library. In order to introd
uce quantum mechanics concepts we develop a series of displays\, activitie
s\, and written resources that build up to quantum entanglement and encour
age people to directly interact with photons in the network. These resourc
es culminate in a projection measurement system located in the library\, w
here the public can choose measurement bases and perform their own CHSH in
equality measurements.\nShort Bio:\nKeshav Kapoor received the B.A. degree
in physics and applied mathematics from the University of California\, Be
rkeley\, in 2021. He is currently pursuing the Ph.D. degree at the Univers
ity of Illinois Urbana–Champaign advised by professors Paul Kwiat and Vi
rginia Lorenz. His research focuses on quantum information science\, optic
s\, and networks. Before starting the PhD program\, he was a Quantum Compu
ting Associate with the Fermi National Accelerator Laboratory (Fermilab)\,
where he had been working on the FQNET Project.\n
URL:https://www.tcs.tifr.res.in/web/events/1471
DTSTART;TZID=Asia/Kolkata:20240902T100000
DTEND;TZID=Asia/Kolkata:20240902T110000
LOCATION:H.B.A. Foyer
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1468
DTSTAMP:20240829T095226Z
SUMMARY:Performativity in Reinforcement Learning
DESCRIPTION:Speaker: Debmalya Mandal (University of Warwick)\n\nAbstract: \
nHow should we design machine learning systems when the underlying environ
ment (e.g. data distribution) changes in response to the deployed model? I
n the context of supervised learning\, the framework of performative predi
ction provides game-theoretic solution concepts that a learner can optimiz
e in the presence of decision-dependent distributions. In this talk\, I wi
ll provide an overview of our work to model such “performativity” in t
he context of reinforcement learning. In particular\, I will describe how
to reach a stable policy in a setting where the underlying MDP reacts to t
he deployed policy. I will end with some open questions\, and if time perm
its\, some of our recent works on reinforcement learning with human feedb
ack.\nShort Bio:\nDebmalya Mandal is an assistant professor at the Univers
ity of Warwick\, UK. He completed his PhD from Harvard University and was
subsequently a postdoc at Columbia University and Max Planck Institute.
He is broadly interested in problems at the interface of machine learni
ng and multi-agent systems.\n
URL:https://www.tcs.tifr.res.in/web/events/1468
DTSTART;TZID=Asia/Kolkata:20240903T100000
DTEND;TZID=Asia/Kolkata:20240903T110000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1454
DTSTAMP:20240826T050725Z
SUMMARY:Markov Decision Processes as Distribution Transformers: Certified P
olicy Verification and Synthesis
DESCRIPTION:Speaker: S. Akshay (Indian Institute of Technology\, Bombay)\n\
nAbstract: \nMarkov decision processes can be viewed as transformers of pr
obability distributions\, giving rise to a sequence of distributions over
MDP states. This view is useful in many applications\, e.g.\, modeling ro
bot swarms or chemical reaction networks\, where the behavior is naturally
interpreted as probability distributions over states. Somewhat surprising
ly\, in this setting\, basic reachability and safety problems turn out to
be computationally intractable. The issue is further complicated by the q
uestion of how much memory is allowed: even for simple examples\, policies
for safety objectives over distributions can require infinite memory and
randomization.\nIn light of this\, we ask what can one do to tackle these
problems in theory and in practice? After taking a look at some theoretica
l insights\, we adopt an over-approximation route to approach these questi
ons. Inspired by the success of invariant synthesis in program verificatio
n\, we develop a framework for inductive template-based synthesis of certi
ficates and policies for safety and reach-avoidance objectives in MDPs. We
show the effectiveness of our approach as well as explore limitations a
nd future perspectives.\n[Based on Joint Work with Krishnendu Chatterjee\,
Tobias Meggendorfer and Djordje Zikelic at CAV'23 and IJCAI'24]\nShort Bi
o:\nS. Akshay is a Professor in the Department of Computer Science and Eng
ineering at IIT Bombay. His research interests span formal methods and AI
with a focus on automata theory\, quantitative (timed/probabilistic) verif
ication and automated synthesis. He has given multiple tutorials on his wo
rk in venues including Highlights 2022\, AAAI 2022 and IJCAI 2022 as well
as helped organize workshops on Automata\, Concurrency and Timed systems i
n India and outside.\n
URL:https://www.tcs.tifr.res.in/web/events/1454
DTSTART;TZID=Asia/Kolkata:20240903T160000
DTEND;TZID=Asia/Kolkata:20240903T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1469
DTSTAMP:20240829T094125Z
SUMMARY:Presburger Arithmetic : Quantifier Elimination and Some Application
s
DESCRIPTION:Speaker: Khushraj Madnani (Max Planck Institute for Software Sy
stems\, Germany)\n\nAbstract: \nIn this talk\, we revisit the fundamental
problem of quantifier elimination in Existential Presburger Arithmetic. As
one of the main highlights\, we challenge the long-standing claim that el
iminating a block of existentially quantified variables necessarily requir
es doubly exponential time. Our recent work refutes this by introducing a
novel procedure which accomplishes quantifier elimination in singly expone
ntial time. The core of our approach is a small model property for paramet
ric integer programming\, which extends the seminal results of von zur Gat
hen and Sieveking on small integer points within convex polytopes. Additio
nally\, if time permits\, I will discuss a compelling application of Presb
urger Arithmetic in proving a dichotomy related to the reachability proble
m for counter machines with infrequent reversals. By analyzing the growth
of small solutions for iterations of Presburger-definable constraints\, we
show that any counter machine falls into one of two categories: (i) the n
umber of reversals is uniformly bounded by a constant across all runs\, or
(ii) the number of reversals grows at least logarithmically with the leng
th of the run. Moreover\, reachability is undecidable for counter machines
where the number of reversals grows logarithmically. This result indicate
s that\, vis-à-vis counter machines\, classical reversal bounding encompa
sses all the decidable cases within the broader framework of infrequent re
versals.\nShort Bio:\nKhushraj Madnani is a postdoctoral researcher at the
Max-Planck Institute for Software Systems in Kaiserslautern\, Germany\,
associated with the Rigorous Software Engineering group and the Models o
f Computation group. His research interests is boradly within the domain o
f formal verification of infinite-state systems\, focusing primarily on (1
) automata and logics for timed systems\, (2) formal logics and models of
computation\, and (3) network controlled cyber physical systems. Khushraj
completed his Master's and Ph.D. in Computer Science and Engineering at th
e Indian Institute of Technology (IIT) Bombay\, Mumbai\, India\, under the
guidance of Prof. S. Krishna and Prof. Paritosh K. Pandya where he defend
ed his thesis titled "On Decidable Extensions of Metric Temporal Logic".
Before joining the Max-Planck Institute\, Khushraj was a postdoctoral rese
archer at the Delft Center for Systems and Control (DCSC) within the Facul
ty of Mechanical Engineering at Delft University of Technology\, The Nethe
rlands. He also served as a visiting postdoctoral fellow at the Tata Insti
tute of Fundamental Research (TIFR) in Mumbai\, India.\n
URL:https://www.tcs.tifr.res.in/web/events/1469
DTSTART;TZID=Asia/Kolkata:20240905T113000
DTEND;TZID=Asia/Kolkata:20240905T123000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1473
DTSTAMP:20240904T102442Z
SUMMARY:An introduction to fairness in reinforcement learning.
DESCRIPTION:Speaker: Vishakha Patil (Indian Institute of Science\, Bangalor
e)\n\nAbstract: \nIn a series of three talks\, we present an overview of w
ork in bandit optimization\, including stochastic and adversarial bandits\
, and will focus on variants of multi-armed bandits where fairness is an a
dditional constraint. In Fair-MAB\, for example\, in addition to the objec
tive of maximizing the sum of expected rewards\, the algorithm also needs
to ensure that at any time\, each arm is pulled at least a pre-specified f
raction of times. We investigate the interplay between learning and fairne
ss in such settings\, and also investigate the cost of fairness.\nShort Bi
o:\nVishakha Patil works in online learning and optimization\, game theory
\, and mechanism design. She completed her PhD from IISc\, supported by a
Google Fellowship and a CII-SERB PMRF. She has also been a Heidelberg Laur
eate Forum Young Researcher\, and her MTech thesis received an Honourable
Mention for the Best Thesis Award at CSA\, IISc.\n
URL:https://www.tcs.tifr.res.in/web/events/1473
DTSTART;TZID=Asia/Kolkata:20240905T140000
DTEND;TZID=Asia/Kolkata:20240905T150000
LOCATION:A-238
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1474
DTSTAMP:20240906T083224Z
SUMMARY:Space-width trade-offs for Resolution
DESCRIPTION:Speaker: Sreejata Kishor Bhattacharya (TIFR)\n\nAbstract: \nWe
consider two natural measures associated with a Resolution refutation*: i
ts width (the widest clause appearing in the refutation) and its space (th
e minimum amount of memory necessary to verify its correctness). A natural
question to ask is the following: can we optimise both parameters at once
? In other words\, can we find a resolution refutation whose space and wid
th are both close to the minimum possible value?I shall exhibit a negative
result in this direction: there exists an unsatisfiable CNF which has a r
efutation with constant space and a refutation with constant width\, but m
aking one parameter small necessarily causes a blowup in the other paramet
er.I shall be following this paper: https://dl.acm.org/doi/10.1145/509907
.509975*PS: Don't worry if you don't know the definition of Resolution ref
utation.\n
URL:https://www.tcs.tifr.res.in/web/events/1474
DTSTART;TZID=Asia/Kolkata:20240906T160000
DTEND;TZID=Asia/Kolkata:20240906T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1466
DTSTAMP:20240830T060930Z
SUMMARY:Fast Algorithms and Data Structures for Regression
DESCRIPTION:Speaker: Deeksha Adil (Institute for Theoretical Studies in ETH
Zurich)\n\nAbstract: \n In this talk I will present state-of-the-art alg
orithms for $\\ell_{\\infty}$-norm regression. Our algorithms involve new
techniques which combine acceleration algorithms with advanced data struct
ures. In order to\ndo so\, we propose a novel acceleration scheme for mult
iplicative weight update algorithms that exhibits stabiliy and robustne
ss\, which are required for the efficient implementations of the inverse m
aintenance data structures. Our work is the first to use acceleration and
inverse maintenance together efficiently\, finally making the two most imp
ortant building blocks of modern structured convex optimization compatible
.\nShort Bio:\nDeeksha Adil has been a Junior Fellow at the Institute for
Theoretical Studies at ETH Zurich since January 2023. She focuses on desig
ning fast algorithms with provable guarantees for problems in optimization
\, machine learning\, and theoretical computer science. She completed a Ph
.D. at the University of Toronto under the supervision of Sushant Sachdeva
in August 2022 after which she was visiting the University of Michigan in
the Fall of 2022. Prior to that\, she earned a BS-MS in mathematics from
the Indian Institute of Science Education and Research\, Pune\, in 2017.\n
URL:https://www.tcs.tifr.res.in/web/events/1466
DTSTART;TZID=Asia/Kolkata:20240909T160000
DTEND;TZID=Asia/Kolkata:20240909T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1429
DTSTAMP:20240910T040613Z
SUMMARY:Fusing AI and Formal Methods for Automated Synthesis
DESCRIPTION:Speaker: Priyanka Golia (IIT\, Delhi)\n\nAbstract: \nWe entrust
large parts of our daily lives to computer systems\, which are becoming i
ncreasingly more complex. Developing scalable yet trustworthy techniques f
or designing and verifying such systems is an important problem. In this t
alk\, our focus will be on automated synthesis\, a technique that uses f
ormal specifications to automatically generate systems (such as functions\
, programs\, or circuits) that provably satisfy the requirements of the sp
ecification. \nWe will introduce a state-of-the-art functional synthesis
algorithm that leverages artificial intelligence to provide an initial gu
ess for the system and then uses formal methods to repair and verify the g
uess to synthesize a system that is correct by construction. We will concl
ude by exploring the potential for combining AI and formal methods to addr
ess real-world scenarios.\n \nShort Bio:\nPriyanka Golia is an Assistant
Professor in the Computer Science and Engineering Department at Indian Ins
titute of Technology Delhi. Prior to that\, She was faculty at CISPA Helmh
oltz Center for Information Security\, Germany. Her research interests lie
at the intersection of formal methods and artificial intelligence. In par
ticular\, her work so far has focused on designing scalable automated synt
hesis and testing techniques. Her work has been awarded Best Paper Nominat
ion at ICCAD-21\, Best Paper Candidate at DATE-23\, Invited for FMSD speci
al issues. She was named one of the EECS Rising Stars in 2022.\n
URL:https://www.tcs.tifr.res.in/web/events/1429
DTSTART;TZID=Asia/Kolkata:20240910T160000
DTEND;TZID=Asia/Kolkata:20240910T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1462
DTSTAMP:20240830T044707Z
SUMMARY:On the composition question for randomized query complexity.
DESCRIPTION:Speaker: Swagato Sanyal (Indian Institute of Technology\, Khara
gpur)\n\nAbstract: \nA query algorithm (also known as a decision tree) tha
t computes a Boolean function f on n variables\, queries various bits of a
n input to f\, possibly in an adaptive fashion and using randomness. It ev
entually produces a bit as an output\, which is supposed to equal the valu
e of f on that input with high probability. The complexity measure of such
an algorithm is the number of queries that it makes in the worst case. Th
is talk considers the query complexity of a class of Boolean function call
ed 'composed functions'. Composition of two Boolean functions f and g is\,
informally speaking\, the Boolean function (say h) obtained by successive
applications of g and f. The composition question\, instantiated for quer
y complexity\, asks whether there exists a query algorithm that computes h
that is significantly more efficient than what the definition of h sugges
ts: first compute g and then compute f. The question has been the centre o
f a lot of research\, and stands open to this day. In this work we present
some results in relation to this question. The talk is based on a work th
at was accepted to STACS 2024.\nShort Bio:\nSwagato Sanyal is an Assistant
Professor at the Department of Computer Science and Engineering\, IIT Kha
ragpur. Prior to this\, he was a post-doctoral research fellow at NTU and
NUS\, Singapore. He obtained is PhD degree from STCS\, TIFR in 2017. He is
interested broadly in computational complexity theory\, and more specific
ally in query and communication complexity and Boolean function analysis.\
n
URL:https://www.tcs.tifr.res.in/web/events/1462
DTSTART;TZID=Asia/Kolkata:20240917T160000
DTEND;TZID=Asia/Kolkata:20240917T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1432
DTSTAMP:20240905T080101Z
SUMMARY:Polynomial Time Algorithms for Integer Programming and Unbounded Su
bset Sum in the Total Regime
DESCRIPTION:Speaker: Divesh Aggarwal (National University of Singapore (NUS
))\n\nAbstract: \nThe Unbounded Subset Sum (USS) problem is an NP-hard com
putational problem where the goal is to decide whether there is a positive
integer combination of numbers a1\,...\,an that is equal to b. The proble
m can be solved in pseudopolynomial time\, while there are specialized ca
ses\, such as when b exceeds the Frobenius number of a1\,...\,an\, for whi
ch a solution is guaranteed to exist. \n \nIn this talk\, I will conside
r the search version of this problem\, where the goal is to find the solut
ion. explores the concept of totality in USS. The challenge in this setti
ng is to actually find a solution\, even though we know its existence is g
uaranteed. We focus on the instances of USS where solutions are guaranteed
for large b. I will show that if b is slightly larger than the Frobenius
number\, then a solution can be found in polynomial time. We then show ho
w our results extend to Integer Programming with Equalities (ILPE)\, highl
ighting conditions under which ILPE becomes total. We investigate the diag
onal Frobenius number\, which is the appropriate generalization of the Fro
benius number to this context. In this setting\, we give a polynomial-time
algorithm to find a solution of ILPE. The bound obtained from our algorit
hmic procedure for finding a solution almost matches the recent existentia
l bound of Bach\, Eisenbrand\, Rothvoss\, and Weismantel (2024).\n \nThis
talk is based on joint work with Antoine Joux\, Miklos Santha\, and Karol
Wegrzycki. \nShort Bio:\nDivesh Aggarwal is an Associate Professor in th
e Department of Computer Science and a Principal Investigator at the Centr
e for Quantum Technologies at the National University of Singapore. He ear
ned his PhD from ETH Zurich\, where he focused on the theoretical foundati
ons of cryptography. His research spans lattice-based cryptography\, pseud
orandomness\, computational complexity\, and coding theory. He has made si
gnificant contributions to the field\, including developing some of the fa
stest known algorithms for lattice problems and advancing the understandin
g of their computational hardness. His work also includes pioneering const
ructions in non-malleable codes and extractors\, which play a crucial role
in modern cryptographic protocols. In recognition of his innovative resea
rch\, he was awarded the NRF Investigatorship in 2024\, a prestigious hono
r granted to leading scientists in Singapore to support high-impact\, risk
-taking research\n
URL:https://www.tcs.tifr.res.in/web/events/1432
DTSTART;TZID=Asia/Kolkata:20240924T160000
DTEND;TZID=Asia/Kolkata:20240924T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1464
DTSTAMP:20240816T072020Z
SUMMARY:To be announced
DESCRIPTION:Speaker: Jayakrishnan Nair (Indian Institute of Technology\, Bo
mbay)\n\nAbstract: \n \nShort Bio:\n \n
URL:https://www.tcs.tifr.res.in/web/events/1464
DTSTART;TZID=Asia/Kolkata:20241001T160000
DTEND;TZID=Asia/Kolkata:20241001T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1475
DTSTAMP:20240909T051225Z
SUMMARY:To be announced
DESCRIPTION:Speaker: Ruta Mehta (University of Illinois at Urbana-Champaign
)\n\nAbstract: \n \nShort Bio:\n \n
URL:https://www.tcs.tifr.res.in/web/events/1475
DTSTART;TZID=Asia/Kolkata:20241011T160000
DTEND;TZID=Asia/Kolkata:20241011T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1479
DTSTAMP:20240912T090252Z
SUMMARY:Transfer Q*: Principled Decoding for LLM Alignment
DESCRIPTION:Speaker: Amrit Singh Bedi (University of Central Florida)\n\nAb
stract: \nTraditional fine-tuning of foundation models is computationally
heavy\, involving updates to billions of parameters. A promising alternati
ve\, alignment via decoding\, adjusts the response distribution directly
without model updates to maximize a target reward r\, thus providing a lig
htweight and adaptable framework for alignment. However\, principled deco
ding methods rely on oracle access to an optimal Q-function (Q*)\, which
is often unavailable in practice. We propose Transfer Q*\, which implicitl
y estimates the optimal value function for a target reward through a basel
ine model aligned with a baseline reward rBL (which can be different from
the target reward). Our approach significantly reduces the sub-optimality
gap observed in prior SoTA methods and demonstrates superior empirical per
formance across key metrics such as coherence\, diversity\, and quality in
extensive tests on several synthetic and real datasets.\nShort Bio:\nAmri
t Singh Bedi is an assistant professor in the Computer Science department
at the University of Central Florida\, Fl\, USA. Before that\, He was a r
esearch assistant professor in the Computer Science Department at the Uni
versity of Maryland\, College Park\, MD\, USA. He obtained his Ph.D. in El
ectrical Engineering from IIT Kanpur\, Kanpur\, India\, in 2018. Following
his doctoral studies\, he worked as a Research Associate within the Compu
tational and Information Sciences Directorate at the US Army Research Labo
ratory (ARL) in Adelphi\, MD\, USA\, from 2019 to 2022. His research inter
ests lie in artificial intelligence (AI) for autonomous systems\, with spe
cific emphasis on scalable & sample-efficient learning algorithms. Current
ly\, he is working on the problem of AI alignment in language models. Hi
s paper was selected as one of the Best Paper Finalists at the 2017 IEEE A
silomar Conference on Signals\, Systems\, and Computers. He received an ho
norable mention from the IEEE Robotics and Automation Letters in 2020. He
was awarded the Amazon Research Award in 2022.\n
URL:https://www.tcs.tifr.res.in/web/events/1479
DTSTART;TZID=Asia/Kolkata:20241022T160000
DTEND;TZID=Asia/Kolkata:20241022T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1478
DTSTAMP:20240909T104205Z
SUMMARY:To be announced
DESCRIPTION:Speaker: Shahar Dobzinski (Weizmann Institute of Science)\n\nAb
stract: \n \nShort Bio:\n \n
URL:https://www.tcs.tifr.res.in/web/events/1478
DTSTART;TZID=Asia/Kolkata:20241119T160000
DTEND;TZID=Asia/Kolkata:20241119T170000
LOCATION:via Zoom in A201
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1477
DTSTAMP:20240911T042624Z
SUMMARY:To be announced
DESCRIPTION:Speaker: Prajakta Nimbhorkar (Chennai Mathematical Institute)\n
\nAbstract: \n \nShort Bio:\nPrajakta Nimbhorkar is an Associate Professo
r at Chennai Mathematical Institute. Her research interests are broadly in
design and analysis of algorithms. She has completed her Ph.D at The Inst
itute of Mathematical Sciences\, Chennai under the supervision of Prof. Me
ena Mahajan.\n
URL:https://www.tcs.tifr.res.in/web/events/1477
DTSTART;TZID=Asia/Kolkata:20241203T160000
DTEND;TZID=Asia/Kolkata:20241203T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VEVENT
UID:www.tcs.tifr.res.in/event/1476
DTSTAMP:20240909T103752Z
SUMMARY:To be announced
DESCRIPTION:Speaker: Aiswarya C (Chennai Mathematical Institute)\n\nAbstrac
t: \n \nShort Bio:\n
URL:https://www.tcs.tifr.res.in/web/events/1476
DTSTART;TZID=Asia/Kolkata:20241210T160000
DTEND;TZID=Asia/Kolkata:20241210T170000
LOCATION:A-201 (STCS Seminar Room)
END:VEVENT
BEGIN:VTIMEZONE
TZID:Asia/Kolkata
BEGIN:STANDARD
DTSTART:20000101T150000
TZNAME:IST
TZOFFSETTO:+0530
TZOFFSETFROM:+0530
END:STANDARD
END:VTIMEZONE
END:VCALENDAR