Abstract

In this work we offer a framework for reasoning about a wide class ofexisting objectives in machine learning. We develop a formal correspondencebetween this work and thermodynamics and discuss its implications.

 

Quick Read (beta)