FSPN: A New Class of Probabilistic Graphical Model

We introduce FSPN, a new class of probabilistic graphical model (PGM). FSPN is designed to overcome the drawbacks in expressiveness and tractability of other PGMs. Specifically, Bayesian network (BN) has low inference efficiency and sum product network performance significantly degrades in presence of highly correlated variables. FSPN absorbs their advantages by adaptively modeling the joint distribution of variables according to their dependence degree. It can simultaneously attain the two desirable goals: high estimation accuracy and fast inference speed. We present efficient probability inference and structure learning algorithms for FSPN, along with theoretical analysis and extensive evaluation evidence. Our experimental results on the benchmark datasets indicate that FSPN is a new SOTA PGM.