摘要
受限玻尔兹曼机是深度学习中的重要模型,以其为基础的卷积受限玻尔兹曼机模型被广泛应用于图像处理与语音识别等领域,但其存在训练时间过长的问题。为此,使用快速持续对比散度(FPCD)算法对卷积受限玻尔兹曼机进行学习,从而提高模型的学习速度和分类精度。实验结果表明,与PCD,CD_1等算法相比,FPCD算法可有效提高卷积受限玻尔兹曼机的分类性能。
Restricted Boltzmann Machine(RBM) is one of the important models in deep learning. The Convolutional RBM(CRBM) model based on RBM is widely used in image processing and speech recognition. However, the long training time is still a problem of the CRBM model that cannot be ignored. In this paper, the Fast Persistent Contrastive Divergence (FPCD) algorithm is used to train the CRBM, to improve the learning speed and classification accuracy of the model. Experimental results show that, compared with PCD, CD 1 and other algorithms, FPCD can improve the classification performance of CRBM.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第9期174-179,共6页
Computer Engineering
基金
国家自然科学基金资助项目(61163036
61163039)
甘肃省科技计划基金资助项目(1606RJZA047)
甘肃省高校基本科研业务费专项基金资助项目
甘肃省高校研究生导师基金资助项目(1201-16)
西北师范大学第三期知识与创新工程科研骨干基金资助项目(nwnu-kjcxgc-03-67)
关键词
卷积受限玻尔兹曼机
深度学习
快速持续对比散度
训练时间
分类精度
Convolutional Restricted Boltzmann Machine (CRBM)
deep learning
Fast Persistent Contrastive Divergence (FPCD)
train time
classification accuracy