随机受限玻尔兹曼机组设计  

Design of Random Restricted Boltzmann Machine Group

作　　者：刘凯 张立民 周立军 LIUKai ZHANGLimin ZHOULijun(Department of Basic Experiment Institute of Information Fusion, Naval Aeronautical University, Yantai 264001, Shandong, China)

机构地区：[1]海军航空大学基础实验部,山东烟台264001 [2]海军航空大学信息融合所,山东烟台264001

出　　处：《上海交通大学学报》2017年第10期1235-1240,共6页Journal of Shanghai Jiaotong University

基　　金：国家自然科学基金(61102167);泰山学者工程专项经费(No.st201511020)资助

摘　　要：为提高受限玻尔兹曼机(Restricted Boltzmann Machine,RBM)数据学习能力和抑制训练的特征同质化问题,提出一种随机受限玻尔兹曼机组(Random-RBM Group,R-RBMG)设计.对观测数据进行随机维度组合,在随机维度组合的基础上构建子RBM群组并实施训练,随后依据神经网络的层数选择模型特征组合方式,针对浅层结构设置为均值组合方式,针对深层模型设置为隐单元叠加方式.理论分析表明,随着组内模型数目的增加,R-RBMG所要学习的训练目标将逐渐接近于标准RBM的训练目标,并且能够有效减少特征同质化带来的影响;实验结果表明,与衰落机制相比,R-RBMG能够有效提高RBM的特征学习能力,应用所组建的浅层结构和深层结构特征,将MNIST(Mixed National Institute of Standards and Technology)数据库实验的分类准确率分别提高了2%和0.4%.To improve the restricted Boltzmann machine （RBM）＇s data generalization ability and resolve the features homogenization problem, a random RBM group （R-RBMG） design is proposed. The dimensions of observation data were randomly divided into groups, and the child-RBMs were built based on the combined data group. Two methods based on the structural stories were used to compose hidden units＇ layer finally, shallow structure by mean output, and deep structure through the formation of high-level hidden units＇ layer： The theoretical analysis shows that, with the increase of models＇ number in the group, the training objectives of R-RBMG will gradually approach the training objectives of standard RBM, and cari effectively reduce the impact of feature homogeneity. The experimental results show that, compared with dropout algorithm, the.proposed R-RBMG can effectively improve the feature learning ability of RBM, and use the shallow structure and deep structure features to increase the classification accura- cy of mixed national institute of standards and technology （MNIST） database experiment by 2% and 0.4%.

关 键 词：机器学习 深度学习 受限玻尔兹曼机 深度玻尔兹曼机 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]