Probabilistic graphical models originated (as "spin systems") in statistical mechanics, but subsequently developed into a popular tool for the modeling of interactions between components of large systems in a wide variety of applications ranging from computer science to biology. An interesting feature of graphical models is that—even in applications that are superficially far away from physics—they often continue to exhibit a sharp change in behavior over a small change of parameters: a "phase transition".
In this talk, we will see examples of such computational and statistical "phase transitions" in applications of graphical models. The examples will be from problems of statistical inference and sampling, and from "community detection", a problem about learning the model itself.
