On Inference and Learning With Probabilistic Generating Circuits

Explore a 22-minute oral presentation from the Uncertainty in Artificial Intelligence conference that delves into the world of Probabilistic Generating Circuits (PGCs). Discover a new inference algorithm that significantly improves computational efficiency by focusing on the highest degree coefficients of polynomials, resulting in linear complexity relative to circuit size. Learn about the advantages of using division-based fast algorithms for determinant-based circuits, eliminating the need for expansion to division-free circuits. Examine the challenges in learning PGCs from data, including the NP-hardness of recognizing valid PGC encodings. Gain insights into potential solutions, such as restricting learning to PGCs composed of moderate-size subcircuits. Access the presentation slides to visualize key concepts and findings in this cutting-edge research on multivariate probability generating polynomials and their applications in artificial intelligence.