Structural Optimization of Deep Belief Network by Evolutionary Computation Methods including Tabu Search

T his paper proposes structural optimization method of a Deep Belief Network (DBN) which consists of multiple Restricted Boltzmann Machines (RBMs) and a single Feedforward Neural Network (FNN) using several kinds of evolutionary computation methods and modularization. The performance, accuracy of data classification or data prediction, should strongly depend on the structure of the network. Concretely, the number of RMBs, the number of nodes in the hidden layer of RMB. The result of the experiments using some benchmarks for image data classification problems by DBN optimized by the proposed method, DBN without any structural optimization, and some other data classification methods indicate that our proposed method defeats other existing classification methods.