We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Condensed Matter > Disordered Systems and Neural Networks

Title: Developing a Maximum-Entropy Restricted Boltzmann Machine with a Quantum Thermodynamics Formalism

Abstract: We propose a Restricted Boltzmann Machine (RBM) neural network using a quantum thermodynamics formalism and the maximization of entropy as the cost function for the optimization problem. We verify the possibility of using an entropy maximization approach for modeling physical systems by applying it to the Inverse Ising Problem, which describes a system of interacting spins. The temperature-dependent behaviors of physical quantities, such as magnetization and Gibbs energy, were calculated from the Ising Hamiltonian whose coefficients were obtained through entropy maximization and were found to be in good agreement to literature results. These results suggest that RBM neural networks using the principle of maximum entropy can be applied to modeling physical systems that can be described by discrete states, including fundamental quantum physics such as topological systems, and biological systems such as the correlated spiking of neurons.
Comments: 13 pages, 3 figures
Subjects: Disordered Systems and Neural Networks (cond-mat.dis-nn); Computational Physics (physics.comp-ph)
Cite as: arXiv:2103.09482 [cond-mat.dis-nn]
  (or arXiv:2103.09482v1 [cond-mat.dis-nn] for this version)

Submission history

From: Timofey Golubev [view email]
[v1] Wed, 17 Mar 2021 07:33:22 GMT (421kb)

Link back to: arXiv, form interface, contact.