Pattern Recognition on the Basis of Boltzmann Machine Model.

In the article the actual problem of increasing the efficiency of solving the pattern recognition problem is considered. It is described a promising approach to solve this problem by the use of artificial neural networks. It is proposed the model of a neural network as the Boltzmann machine. As a neural network learning algorithm the authors propose to use a simulated annealing algorithm. The deep learning methods of neural networks are considered. The algorithm of neural network functioning based on the Boltzmann machine model is suggested. The authors describe possibilities of using multi-layer neural network models, such as the deep Boltzmann machines. Advantages and disadvantages of the proposed approaches were found out. To estimate the proposed method the authors carried out the comparison of the known test set of sample images (MNIST). The results confirm the effectiveness of the proposed approaches.