一种新的分类受限玻尔兹曼机改进模型  

Improved Model Based on Classification Restricted Boltzmann Machine

作　　者：尹静[1] 李唯唯[1] 杨德红[1] 闫河[1] YIN Jing;LI Wei-wei;YANG De-hong;YAN He(College of Computer Science and Eechnology, Chongqing University of Technology, Chongqing 400054, Chin)

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《小型微型计算机系统》2018年第7期1415-1419,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61173184)资助

摘　　要：分类受限玻尔兹曼机(classification restricted boltzmann machine,ClassRBM)在各种分类问题中得到了广泛应用.ClassRBM是一种自带标签信息的神经网络模型,它使用一个神经元标识某类数据的类标.标签神经元总是稀疏的,一个神经元仅能为网络模型参数提供有限的信息.论文在ClassRBM现有的网络结构上,增加标签神经元个数,使每个类标用K个神经元标识,为网络模型参数提供更多的信息,提升模型表达能力,进而改善ClassRBM的分类性能.论文在不同数据集上进行了测试,结果表明改进模型的分类效果可以优于ClassRBM.The classification restricted boltzmann machine（ ClassRBM） is widely used in various classification applications. It is a type of self-contained neural network model and use a simple neuron to identify a class of data in the label layer. Label neurons are often scarce,such a neuron can only provide limited information for the model parameters. In this paper,Multiple neurons are increased in the label layer on the basis of ClassRBM,and each class is identified by K neurons which provides more information for the model parameters.The improved model is used to improve the classification performance of ClassRBM. The experimental results show that the proposed model can be better than ClassRBM.

关 键 词：分类受限玻尔兹曼机 特征学习 分类 改进模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]