IMPACT semainar: Milan Pepež – „Probabilistic Graph Circuits: Deep Generative Models for Tractable Probabilistic Inference over Graphs“

Datum / čas
Date(s) - 04.02.
11:00 - 12:00

Kategorie ne Kategorie


Dear colleagues,

 

It is our pleasure to invite you to the following seminar organized by the IMPACT group and the Applied Algebra and Geometry (AAG) group:

 

Probabilistic Graph Circuits: Deep Generative Models for Tractable Probabilistic Inference over Graphs

Milan Papež

Czech Technical University in Prague

 

Tuesday 2025-02-04 at 11:00

CIIRC Room B-670 (building B, floor 6)

http://impact.ciirc.cvut.cz/seminars/#seminar2025-02-04-papez

 

Abstract:

Deep generative models (DGMs) have recently demonstrated remarkable success in capturing complex probability distributions over graphs. Although their excellent performance is attributed to powerful and scalable deep neural networks, it is, at the same time, exactly the presence of these highly non-linear transformations what makes DGMs intractable. Indeed, despite representing probability distributions, intractable DGMs deny probabilistic foundations by their inability to answer even the most basic inference queries without approximations or design choices specific to a very narrow range of queries. To address this limitation, we propose probabilistic graph circuits (PGCs), a framework of tractable DGMs that provide exact and efficient probabilistic inference over (arbitrary parts of) graphs. Nonetheless, achieving both exactness and efficiency poses challenges in the permutation-invariant setting of graphs. We demonstrate that these two requirements are contradictory for certain PGC constructions and discuss the consequences for the standard inference queries. To ensure permutation invariance, we design a PGC-based model that sorts input graphs into their canonical ordering. Our experimental results demonstrate that these order-conditioned PGCs can generate novel and chemically valid molecular graphs, performing competitively with, and sometimes surpassing, existing intractable DGMs.

 

For more information on IMPACT seminars, please visit

http://impact.ciirc.cvut.cz/seminars

 

Kind regards,

 

Josef Sivic and Tomas Pajdla

josef.sivic@cvut.cz, pajdla@cvut.cz

IMPACT ( impact.ciirc.cvut.cz )

AAG ( aag.ciirc.cvut.cz )

Czech Institute of Robotics, Informatics and Cybernetics ( www.ciirc.cvut.cz )