Fuzzy Restricted Boltzmann Machines

Restricted Boltzmann Machines are a reconstructive neural network. They derive an implicitly probabilistic model of data which can be used to reconstruct or filter missing data as well as to classify data. This paper develops a deterministic training algorithm and shows how to use that algorithm to automatically derive fuzzy membership classes. The algorithm developed in this paper combines many of the best features of fuzzy learning algorithms and Restricted Boltzmann machines.