Geometry-Aware Probabilistic Circuits via Voronoi Tessellations

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Sidheekh, Sahil, Natarajan, Sriraam
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918533076615168
author Sidheekh, Sahil
Natarajan, Sriraam
author_facet Sidheekh, Sahil
Natarajan, Sriraam
contents Probabilistic circuits (PCs) enable exact and tractable inference but employ data independent mixture weights that limit their ability to capture local geometry of the data manifold. We propose Voronoi tessellations (VT) as a natural way to incorporate geometric structure directly into the sum nodes of a PC. However, naïvely introducing such structure breaks tractability. We formalize this incompatibility and develop two complementary solutions: (1) an approximate inference framework that provides guaranteed lower and upper bounds for inference, and (2) a structural condition for VT under which exact tractable inference is recovered. Finally, we introduce a differentiable relaxation for VT that enables gradient-based learning and empirically validate the resulting approach on standard density estimation tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2603_11946
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Geometry-Aware Probabilistic Circuits via Voronoi Tessellations
Sidheekh, Sahil
Natarajan, Sriraam
Machine Learning
Artificial Intelligence
Probabilistic circuits (PCs) enable exact and tractable inference but employ data independent mixture weights that limit their ability to capture local geometry of the data manifold. We propose Voronoi tessellations (VT) as a natural way to incorporate geometric structure directly into the sum nodes of a PC. However, naïvely introducing such structure breaks tractability. We formalize this incompatibility and develop two complementary solutions: (1) an approximate inference framework that provides guaranteed lower and upper bounds for inference, and (2) a structural condition for VT under which exact tractable inference is recovered. Finally, we introduce a differentiable relaxation for VT that enables gradient-based learning and empirically validate the resulting approach on standard density estimation tasks.
title Geometry-Aware Probabilistic Circuits via Voronoi Tessellations
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2603.11946