Polyhedral approaches to learning Bayesian networks

This talk will cover descriptions of probabilistic conditional independence (CI) models and learning graphical models which has applications in biology (epistasis, gene regulatory networks, protein signaling, systems biology), Markov random processes, probabilistic reasoning, artificial intelligence and more. Given observed data, the goal is to find the CI structure which best explains the data. I will motivate the problem with some interesting biological applications. Next, I will quickly overview graphical approaches to the description of CI structures. Then, I will describe an algebraic description of CI structures introduced by Studeny et al. which has many elegant properties, suitable for applications of linear programming methods. The remainder of the talk will be devoted to linear optimization approaches to learning Bayesian networks, which are special graphical models.

Event Topic

Nonlinear Algebra and Statistics (NLASTATS)