深度信念网络研究进展

深度信念网络（Deep Belief Networks，DBN）作为深度学习（Deep Learning，DL）中的重要模型，目前已被成功应用于人脸识别、手写字体识别、医学图像分析处理等诸多领域。从深度信念网络出发，主要做了四个方面的工作：第一，从受限玻尔兹曼机以及深度信念网络的网络结构和学习过程两个方面阐述了深度信念网络的基本原理；第二，从网络结构和学习算法两个方面总结了深度信念网络的研究进展：在网络结构方面，从网络深度、RBM结构和DBN级联三个角度进行归纳；在学习算法方面，从基本算法、优化算法和融合方法三个方面进行梳理；第三，对深度信念网络在医学图像分析领域中的应用进行了总结；第四，总结了目前DBN存在的问题。

Survey of Deep Belief Network

As an important model in Deep Learning(DL), Deep Belief Network(DBN)has been successfully applied in face recognition, fingerprint recognition, speech recognition, handwriting recognition, medical image analysis and so on.This paper mainly does four aspects of work about deep belief networks. Firstly, this paper discusses the basic principle of depth belief network from two aspects of restricted Boltzmann machine and depth belief network structure and learning process. Secondly, it summarizes the research progress of depth belief network from two aspects of network structure and learning algorithm. In terms of the network structure, it sums up the process from the three aspects:network depth, RBM structure and DBN cascade. In the learning algorithm, it analyzes the learning process from basic algorithm, optimization algorithm and fusion method. Thirdly, it summarizes the application of deep belief network in medical image analysis.Fourth, it summarizes the existing problems of DBN.