Poster

Probabilistic Generating Circuits

Honghua Zhang · Brendan Juba · Guy Van den Broeck

Keywords: [ Graphical Models ]

[ Abstract ]
[ Paper ] [ Visit Poster at Spot D5 in Virtual World ] [ Visit Poster at Spot D4 in Virtual World ]
Thu 22 Jul 9 p.m. PDT — 11 p.m. PDT
 
Oral presentation: Probabilistic Methods 4
Thu 22 Jul 5 p.m. PDT — 6 p.m. PDT

Abstract:

Generating functions, which are widely used in combinatorics and probability theory, encode function values into the coefficients of a polynomial. In this paper, we explore their use as a tractable probabilistic model, and propose probabilistic generating circuits (PGCs) for their efficient representation. PGCs are strictly more expressive efficient than many existing tractable probabilistic models, including determinantal point processes (DPPs), probabilistic circuits (PCs) such as sum-product networks, and tractable graphical models. We contend that PGCs are not just a theoretical framework that unifies vastly different existing models, but also show great potential in modeling realistic data. We exhibit a simple class of PGCs that are not trivially subsumed by simple combinations of PCs and DPPs, and obtain competitive performance on a suite of density estimation benchmarks. We also highlight PGCs' connection to the theory of strongly Rayleigh distributions.

Chat is not available.