In this talk, we would discuss the inference and learning problem on Markov random fields. The key role of inference is motivated by the need for answering general queries and the need for model learning. We start by recapping the classical deterministic methods, e.g., mean-field, belief propagation, generalized belief propagation. Afterward, we introduce our recent development on this track by bringing in neural networks into the problem. The intuition and merits of these methods would be discussed.
This talk will be given over zoom: https://kth-se.zoom.us/j/63467923177