Boltzmann machine
Stochastic neural network with binary units and an energy function from statistical physics; learns hidden representations but typically needs slow Markov-chain (MCMC) sampling to train.
A Boltzmann machine is a neural network of stochastic binary units connected symmetrically. Its states follow an energy function and the Boltzmann distribution from statistical physics. David Ackley, Geoffrey Hinton, and Terrence Sejnowski published a learning algorithm in 1985; the model builds on John Hopfield’s associative networks and can learn hidden representations in unsupervised learning.
Training usually requires lengthy MCMC sampling based on Markov chains, which is slow at generation. A restricted Boltzmann machine (RBM) limits connections between units in the same layer. Geoffrey Hinton introduced contrastive divergence in 2002 as a faster training approximation that made RBMs practical and later enabled deep belief networks. See the article and timeline entry.