Marginal Determinism in Structured Decomposable Circuits

Explore a comprehensive lecture on marginal determinism in structured decomposable circuits presented by Benjie Wang from the University of Oxford. Delve into the world of probabilistic circuits (PCs) and their efficient inference routines. Learn about the novel concept of md-vtrees, a structural formulation of marginal determinism in structured decomposable PCs that generalizes previously proposed classes like probabilistic sentential decision diagrams. Discover how md-vtrees can be utilized to derive tractability conditions and efficient algorithms for compositional inference queries. Gain insights into the first polytime algorithms for causal inference queries, such as backdoor adjustment on PCs. This 43-minute talk, part of the Probabilistic Circuits and Logic series at the Simons Institute, offers a deep dive into advanced concepts in probabilistic modeling and inference.

Marginal Determinism in Structured Decomposable Circuits