Boltzmann Machine

CORE DEFINITION

The Boltzmann machine is a type of stochastic neural network and recurrent neural network, invented in 1985 by Canadian computer scientist Geoffrey Hinton and American computer scientist Terry Sejnowski. It can be viewed as a stochastic, generative counterpart of the Hopfield neural network. It was one of the first neural networks capable of learning internal representations and able to represent and (given sufficient time) solve complex combinatorial optimization problems. However, Boltzmann machines without specific restrictions on connectivity have not yet been proven useful for practical problems in machine learning. Thus, they are currently only of theoretical interest. Nevertheless, due to the local and Hebbian nature of the training algorithm, and their parallelism similar to simple physical processes, if the connectivity is constrained (i.e., restricted Boltzmann machines), the learning method can be efficient enough for practical problems. It is named after the Boltzmann distribution, which is used in the sampling function of the Boltzmann machine.