Bayesian Learning of Probabilistic Circuits with Domain Constraints

Athresh Karanam, Saurabh Mathur, Sahil Sidheekh, Sriraam Natarajan

Keywords: Knowledge-intensive learning, Probabilistic Circuits

TL;DR: We propose a novel unified framework for knowledge-intensive structure and parameter learning of PCs.

Abstract: Probabilistic Circuits (PCs) have emerged as an efficient framework for representing and learning complex probability distributions. However, existing research on PCs primarily focuses on data-driven parameter learning, with limited exploration of knowledge-intensive learning and structure learning. In this work, we propose to address these gaps by introducing a comprehensive approach to incorporating various kinds of domain knowledge into the learning of a PC's structure as well as its parameters.

Submission Number: 20