Boltzmann Machines

The purpose of this paper is to introduce the reader to a specific type of neural networks called Boltzmann Machines. The Boltzmann Machine is a model originally proposed in a paper of Hinton and Sejnowski (1983) (see also Ackley, Hinton and Sejnowski (1985)), and it possesses attractive properties for solving problems from such diverse areas as: pat tern recognition, combinatorial optimization and learning. The combination of massive parallelism from neural computing and simulated annealing is the characteristic feature of Boltzmann Machines and results in a promising computational tool. Potential benefits of the Boltzmann Machine entail among others: