A Dynamic Deep Belief Network Analysis Method

Abstract

The invention relates to a dynamic network structure training analysis method based on a deep belief network in the field of deep learning. The purpose of the present invention is to overcome several deficiencies of the deep belief network, and provide a dynamic network structure training analysis strategy from the whole to the special one based on the deep belief network. This strategy refers to the analysis strategy of the brain when human beings recognize objects, and introduces the process of detailed analysis. This strategy is aimed at classification problems, and two types of networks are generated during the training phase, including a global network and several special networks for specific categories. In the prediction stage, the output results of the two networks are considered comprehensively, so that the prediction of the entire model is more targeted, thereby improving the accuracy of the classification model.

Description

technical field [0001] The invention relates to a dynamic network structure training analysis method based on a deep belief network in the field of deep learning. [0002] technical background [0003] ●Deep Belief Network [0004] Deep belief network (Deep belief network, DBN) is a probabilistic generation model in the field of deep learning. This model was proposed by Geoffrey Hinton in his paper "A Fast Learning Algorithm For Deep Belief Nets" published in 2006. It is usually used for classification . As opposed to traditional neural networks for discriminative models, deep belief networks are used to build a joint distribution between observations and labels. [0005] Theoretically, for the neural network structure, the more layers of hidden layers, the stronger the expressive power of the model. However, when the number of network layers gradually deepens, using the traditional gradient descent algorithm to optimize parameters will cause some problems. For example, w...

AI Technical Summary

Problems solved by technology

[0008] The deep belief network has several obvious shortcomings: 1. When training the model, it strives to find the comprehensive optimal fit of all the data in the entire training set, without considering some special classes; 2. After the entire model is trained, it will not be changed. , all the test set data are input into a model for prediction, which is convenient but lacks certain pertinence

Images