FSPN: A New Class of Probabilistic Graphical Model

TL;DR

FSPNs are a novel probabilistic graphical model that adaptively balances estimation accuracy and inference speed, outperforming existing models like Bayesian networks and sum product networks in various datasets.

Contribution

Introduction of FSPNs, a new class of PGMs that improve inference efficiency and accuracy by adaptively modeling variable dependencies based on their correlation levels.

Findings

FSPNs achieve higher estimation accuracy than traditional PGMs.

FSPNs provide faster inference compared to Bayesian networks and SPNs.

Experimental results demonstrate FSPNs' superiority on synthetic and benchmark datasets.

Abstract

We introduce factorize sum split product networks (FSPNs), a new class of probabilistic graphical models (PGMs). FSPNs are designed to overcome the drawbacks of existing PGMs in terms of estimation accuracy and inference efficiency. Specifically, Bayesian networks (BNs) have low inference speed and performance of tree structured sum product networks(SPNs) significantly degrades in presence of highly correlated variables. FSPNs absorb their advantages by adaptively modeling the joint distribution of variables according to their dependence degree, so that one can simultaneously attain the two desirable goals: high estimation accuracy and fast inference speed. We present efficient probability inference and structure learning algorithms for FSPNs, along with a theoretical analysis and extensive evaluation evidence. Our experimental results on synthetic and benchmark datasets indicate the superiority of FSPN over other PGMs.