A removing redundancy Restricted Boltzmann Machine
Yun Jiang, Jize Xiao, Xi Liu, Jinquan Hou · 2018
As one of the most important models of deep learning, Restricted Boltzmann Machine(RMB) has been extensively studied and concerned in recent years. However, because RBM model has high redundancy, it greatly affects its convergence speed and training time. In this paper, a Removing Redundancy Restricted Boltzmann Machine is proposed to improve the original RBM from the viewpoint of eliminating redundant hidden units. The structure of the Restricted Boltzmann Machine is optimized by removing redundancy, which greatly reduces the number of hidden units, improves the training efficiency of the RBM and shortens the training time and improves the convergence rate of the model. Experimental results show, the improved method can eliminate a large number of redundant hidden elements without affecting the Reconstruction Error, which is very important for optimizing the structure of the restricted Boltzmann machine and accelerating the learning speed.