Optimization of DBN Network Structure Based on Information Entropy
Liao Qiang;Zhang Jie;College of Communication Engineering, Chengdu University of Information Technology;
In order to solve the problem that the deep depth of the depth network(DBN) is difficult to extract the appropriate depth of the network and the number of hidden neurons, the traditional reconstruction error is used to calculate and judge the depth of the network. From the perspective of information expression, Based on the relationship between information entropy and hidden layer, an optimization method based on information entropy to determine the number of hidden neurons is proposed, and the DBN network model is established. The information entropy and the input layer and hidden layer are analyzed. The structure tends to be better The experimental results on handwritten numeral recognition show that the method can self-organize the depth of the network and the number of hidden neurons. Compared with the test data of the network depth according to the reconstruction error method, the average error is reduced by 1.47%, which effectively optimizes the DBN network structure, reduces the training time of the network, improves the network precision and the recognition accuracy.
【CateGory Index】： TP183