In our recent #AAAI25 work, we propose a unified framework for expressing domain knowledge about variables like monotonicity & independence. We use this framework to learn complex Probabilistic Circuit models from small & noisy data sets by exploiting multiple forms of knowledge.
Glad I had the opportunity to present our work on explaining sum-product networks using a tree-structured representation of context-specific independencies encoded by the SPN. Thanks to @mathursaurabh96, Dr. David Haas, @pedjagogue, @kerstingAIML and my advisor @Sriraam_UTD.
Bhagirath Athresh Karanam, Saurabh Mathur, Predrag Radivojac, David M Haas, Kristian Kersting and Sriraam Natarajan. Explaining Deep Tractable Probabilistic Models: The sum-product network case. #PGM2022