Organizer: Nicholas Ruozzi
Office: INR 136
Phone: 37551
Email: nicholas.ruozzi@epfl.ch
Meetings: Fridays 4:15pm – 5:30pm in room INR 113
Overview
The objective of the reading group is to understand the recent results in the study of graphical models and approximate inference. No background in graphical models or message passing is required.
A tentative list of topics includes:

Factor graphs and message passing as dynamic programming on a tree

Message passing (belief propagation, maxproduct, minsum): admissibility, consistency, and computation trees

Message passing for the maximum weight matching problem

Message passing for optimization: MAP LP and reparameterizations

Graph covers and pseudocodewords

Convergent and correct message passing: TRMP, MPLP, etc.

Variational approximations and belief propagation

Variational approximations: zero temperature limits of BP and generalized belief propagation

Convex entropy approximations and convergent message passing: TRBP and others

Graph covers and the Bethe partition function

Loop Expansions
Schedule