Machine-learning iterative calculation of entropy for physical systems

Abstract

The calculation of entropy of a physical system is a fundamental step in learning its thermodynamic behavior. However, current methods to compute the entropy are often system specific and computationally costly. Here, we propose a method that is efficient, accurate, and general for computing the entropy of arbitrary physical systems. Our method is based on computing the mutual information between subsystems within the studied system, using a convolutional neural network. This iterative procedure provides accurate entropy evaluation for systems in and out of equilibrium.

Document Details

Document Type
Pub Defense Publication
Publication Date
Nov 19, 2020
Source ID
10.1073/pnas.2017042117

Entities

People

  • Amit Nir
  • Eran Sela
  • Roy Beck-Barkai
  • Yohai Bar-Sinai

Organizations

  • Army Research Office
  • Google Research
  • Israel Science Foundation
  • Tel Aviv University
  • United States – Israel Binational Science Foundation

Tags

Fields of Study

  • Physics

Readers

  • Fluid Dynamics.
  • Neural Network Machine Learning.
  • Software Engineering.

Technology Areas

  • AI & ML
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks