TherML: The Thermodynamics of Machine Learning

Alexander A. Alemi, Ian Fischer

Abstract: In this work we offer an information-theoretic framework for representation learning that connects with a wide class of existing objectives in machine learning. We develop a formal correspondence between this work and thermodynamics and discuss its implications.

Keywords: representation learning, information theory, information bottleneck, thermodynamics, predictive information

TL;DR: We offer a framework for representation learning that connects with a wide class of existing objectives and is analogous to thermodynamics.