Algorithms for optimization problems such as selection, ranking, and classification typically assume that the inputs are what they are promised to be. However, in several real-world applications of these problems, the input may contain systematic biases along socially salient attributes associated with inputs such as race, gender, or political opinion. Such biases can not only lead the outputs of the current algorithms to output sub-optimal solutions with respect to true inputs but may also adversely affect opportunities for individuals in disadvantaged socially salient groups. This talk will consider the question of using optimization to solve the aforementioned problems in the presence of biased inputs. It will start with models of biases in inputs and discuss alternate ways to design algorithms for the underlying problem that can mitigate the effects of biases by taking into account knowledge about biases.
This talk is based on several joint works with Elisa Celis.
Short Bio: Nisheeth Vishnoi is the A. Bartlett Giamatti Professor of Computer Science at Yale University. His research lies at the intersection of computer science, mathematics, and societal challenges in an increasingly algorithmic world, with a focus on the foundations and ethics of artificial intelligence. He is a Fellow of the Association for Computing Machinery (ACM) and the American Mathematical Society (AMS), recognized for his contributions to theoretical computer science and its connections to mathematics, sciences, and social sciences.
