Nonequilibrium Thermodynamics of Restricted Boltzmann Machines
arXiv:1704.08724 · doi:10.1103/PhysRevE.96.022131
Abstract
In this work, we analyze the nonequilibrium thermodynamics of a class of neural networks known as Restricted Boltzmann Machines (RBMs) in the context of unsupervised learning. We show how the network is described as a discrete Markov process and how the detailed balance condition and the Maxwell-Boltzmann equilibrium distribution are sufficient conditions for a complete thermodynamics description, including nonequilibrium fluctuation theorems. Numerical simulations in a fully trained RBM are performed and the heat exchange fluctuation theorem is verified with excellent agreement to the theory. We observe how the contrastive divergence functional, mostly used in unsupervised learning of RBMs, is closely related to nonequilibrium thermodynamic quantities. We also use the framework to interpret the estimation of the partition function of RBMs with the Annealed Importance Sampling method from a thermodynamics standpoint. Finally, we argue that unsupervised learning of RBMs is equivalent to a work protocol in a system driven by the laws of thermodynamics in the absence of labeled data.
arXiv comment10 pages, 2 figures
References in corpus (11)
- A nonequilibrium equality for free energy differences
- Stochastic thermodynamics, fluctuation theorems, and molecular machines
- The Entropy Production Fluctuation Theorem and the Nonequilibrium Work Relation for Free Energy Differences
- Searching for Exotic Particles in High-Energy Physics with Deep Learning
- Stochastic thermodynamics: Principles and perspectives
- Fluctuation theorems for stochastic dynamics
- Classical and Quantum Fluctuation Theorems for Heat Exchange
- Learning Thermodynamics with Boltzmann Machines
- Enhanced Higgs to Searches with Deep Learning
- Stochastic Thermodynamics of Learning
- Discrete fluctuations in memory erasure without energy cost
Cited by in corpus (11)
- Scale-invariant Feature Extraction of Neural Network and Renormalization Group Flow
- Thermodynamics of Restricted Boltzmann Machines and related learning dynamics
- Minimal model of permutation symmetry in unsupervised learning
- Heat distribution in open quantum systems with maximum entropy production
- Thermodynamic efficiency of learning a rule in neural networks
- Thermodynamics of the Ising model encoded in restricted Boltzmann machines
- The principles of adaptation in organisms and machines I: machine learning, information theory, and thermodynamics
- Revisiting the dynamics of Bose-Einstein condensates in a double well by deep learning with a hybrid network
- Nonequilibrium thermodynamics of self-supervised learning
- Rebooting Neuromorphic Hardware Design -- A Complexity Engineering Approach
- Emergence of Nonequilibrium Latent Cycles in Unsupervised Generative Modeling