Graphical Models, Exchangeable Models and Graphons Workshop

Graphical model (GM) is a succinct way to represent a complex probability distribution on a collection of random variables. Such a probability distribution factorizes according to the graph underlying the GM. As such, the graph structure of the GM encodes interdependencies among the variables, making GMs a powerful and general framework within which to reason about high-dimensional data. The combinatorial structure brings to the fore the computational aspect underlying statistical tasks, helping to design various simple-to-implement and scalable statistical inference algorithms. As a result, GM are extensively used in practice across domains including communication (e.g. LDPC codes), control (e.g. Kalman filters), social science (e.g. denoising surveys), natural language processing (e.g. topic model ), signal processing (e.g. speech recognition), image processing (e.g. deep graphical model), machine learning (e.g. recommendation systems), biology and sciences at large (gene regulatory network, causal inference). On the other hand, it has driven exciting intellectual quest of understanding the boundary of computational and statistical tradeoffs as well as efficient inference algorithms. detection. The workshop will be preceded by a one-day bootcamp on Sunday, August 18, with the goal of presenting the basic techniques, definitions and goals in several of the communities.

For more information please go to http://mifods.mit.edu/graphical.php.