Probabilistic Flow Circuits: Towards Deep Models for Tractable Inference - Oral Session 6

Explore a 27-minute conference talk from the Uncertainty in Artificial Intelligence (UAI) 2023 Oral Session 6 that introduces Probabilistic Flow Circuits, a novel approach to increasing the expressivity of probabilistic circuits. Delve into the theoretical foundations and practical implications of combining probabilistic circuits with normalizing flows to create more powerful yet tractable probabilistic models. Learn how the researchers establish the requirement of decomposability for maintaining tractability in these hybrid models. Examine the empirical evaluation that demonstrates the enhanced expressivity and tractability of Probabilistic Flow Circuits, which extend traditional circuits by incorporating normalizing flows at the leaf nodes. Gain insights into this innovative approach to unified deep models for tractable probabilistic inference, presented by Sahil Sidheekh, Kristian Kersting, and Sriraam Natarajan.