Learning the Ising model with generative neural networks
Francesco D’Angelo, Lucas Böttcher
Abstract
This paper studies the representational properties of restricted Boltzmann machines and variational autoencoders in terms of their ability to capture physical features of the Ising model. The authors provide a detailed analysis of different network architectures and training algorithms, and identify significant differences in the learning performance of both probabilistic models
Topics & Concepts
Boltzmann machineIsing modelGenerative grammarArtificial intelligenceRestricted Boltzmann machineArtificial neural networkComputer scienceProbabilistic logicMachine learningGenerative modelTheoretical computer scienceStatistical physicsPhysicsGenerative Adversarial Networks and Image SynthesisModel Reduction and Neural NetworksNeural Networks and Applications