Converting GMMs ↔ PEFs for fast KLD approximation

FrnkNlsn · x · 2026-08-24

Introduces two types of universal density approximators: Gaussian Mixtures Models (GMMs) and Exponential Families with polynomial sufficient statistics (PEFs).

The post links to methods for converting between GMMs and PEFs, enabling fast approximation of Kullback-Leibler divergences (KLDs) between GMMs. Authored by Frank Nielsen of Sony CSL, focusing on information geometry and ML.

Original post →

More from Research

Research channel →