2.845

2023影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于自适应学习率的深度信念网设计与应用

乔俊飞 王功明 李晓理 韩红桂 柴伟

乔俊飞, 王功明, 李晓理, 韩红桂, 柴伟. 基于自适应学习率的深度信念网设计与应用. 自动化学报, 2017, 43(8): 1339-1349. doi: 10.16383/j.aas.2017.c160389
引用本文: 乔俊飞, 王功明, 李晓理, 韩红桂, 柴伟. 基于自适应学习率的深度信念网设计与应用. 自动化学报, 2017, 43(8): 1339-1349. doi: 10.16383/j.aas.2017.c160389
QIAO Jun-Fei, WANG Gong-Ming, LI Xiao-Li, HAN Hong-Gui, CHAI Wei. Design and Application of Deep Belief Network with Adaptive Learning Rate. ACTA AUTOMATICA SINICA, 2017, 43(8): 1339-1349. doi: 10.16383/j.aas.2017.c160389
Citation: QIAO Jun-Fei, WANG Gong-Ming, LI Xiao-Li, HAN Hong-Gui, CHAI Wei. Design and Application of Deep Belief Network with Adaptive Learning Rate. ACTA AUTOMATICA SINICA, 2017, 43(8): 1339-1349. doi: 10.16383/j.aas.2017.c160389

基于自适应学习率的深度信念网设计与应用

doi: 10.16383/j.aas.2017.c160389
基金项目: 

国家杰出青年科学基金 61225016

国家自然科学基金 61533002

国家自然科学基金 61473034

详细信息
    作者简介:

    乔俊飞    北京工业大学教授.主要研究方向为智能控制, 神经网络分析与设计.E-mail:junfeq@bjut.edu.cn

    李晓理    北京工业大学教授.1997年获得大连理工大学控制理论与工程硕士学位, 2000年获得东北大学博士学位.主要研究方向为多模型自适应控制, 神经网络控制.E-mail:lixiaolibjut@bjut.edu.cn

    韩红桂    北京工业大学教授.主要研究方向为污水处理工艺复杂建模与控制, 神经网络分析与设计.E-mail:rechardhan@sina.com

    柴伟    北京工业大学讲师.主要研究方向为系统辨识和状态估计研究.E-mail:chaiwei@bjut.edu.cn

    通讯作者:

    王功明    北京工业大学博士研究生.主要研究方向为深度学习, 神经网络结构设计和优化.本文通信作者.E-mail:xiaowangqsd@163.com

Design and Application of Deep Belief Network with Adaptive Learning Rate

Funds: 

National Natural Science Fund for Distinguished Young Scholars 61225016

National Natural Science Foundation of China 61533002

National Natural Science Foundation of China 61473034

More Information
    Author Bio:

       Professor at Faculty of Information Technology, Beijing University of Technology. His research interest covers intelligent control, analysis and design of neural networks.E-mail:

       Professor at Faculty of Information Technology, Beijing University of Technology. He received his master degree in control theory and control engineering from Dalian University of Technology in 1997, and Ph. D. degree from Northeastern University in 2000, respectively. His research interest covers multiple model adaptive control and neural network control.E-mail:

       Professor at Faculty of Information Technology, Beijing University of Technology. His research interest covers modelling and control in waste water treatment process, analysis and design of neural networks.E-mail:

       Lecturer at Faculty of Information Technology, Beijing University of Technology. His research interest covers system identiflcation and state estimation.E-mail:

    Corresponding author: WANG Gong-Ming    Ph. D. candidate at Faculty of Information Technology, Beijing University of Technology. His research interest covers deep learning, analysis and design of neural networks. Corresponding author of this paper.E-mail:xiaowangqsd@163.com
  • 摘要: 针对深度信念网(Deep belief network,DBN)预训练耗时长的问题,提出了一种基于自适应学习率的DBN(Adaptive learning rate DBN,ALRDBN).ALRDBN将自适应学习率引入到对比差度(Contrastive divergence,CD)算法中,通过自动调整学习步长来提高CD算法的收敛速度.然后设计基于自适应学习率的权值训练方法,通过网络性能分析给出学习率变化系数的范围.最后,通过一系列的实验对所设计的ALRDBN进行测试,仿真实验结果表明,ALRDBN的收敛速度得到了提高且预测精度也有所改善.
    1)  本文责任编委 王占山
  • 图  1  ALRDBN结构

    Fig.  1  The structure of ALRDBN

    图  2  受限玻尔兹曼机

    Fig.  2  Restricted Boltzmann machine

    图  3  CD-k算法

    Fig.  3  he algorithm of CD-k

    图  4  ALRDBN分层表述结构

    Fig.  4  Hierarchical representation structure of ALRDBN

    图  5  ALRDBN顶层的反传误差

    Fig.  5  Error back-propagated from top layer of ALRDBN

    图  6  顶层RBM的重构误差

    Fig.  6  The reconstruction error of top RBM

    图  7  ALRDBN错误识别原图像

    Fig.  7  The original images with classification mistakes of ALRDBN

    图  8  ALRDBN错误识别图像

    Fig.  8  The images with classification mistakes of ALRDBN

    图  9  隐含层神经元数对收敛时间的影响

    Fig.  9  Effect of the number of hidden neurons on convergence time

    图  10  $\alpha$和$\beta$对收敛时间的影响

    Fig.  10  Influence of $\alpha$ and $\beta$ on convergence time

    图  11  ALRDBN训练结果

    Fig.  11  The training results of ALRDBN

    图  12  ALRDBN测试结果

    Fig.  12  The test results of ALRDBN

    图  13  ALRDBN训练RMSE

    Fig.  13  The training RMSE of ALRDBN

    图  14  隐含层神经元数对收敛时间的影响

    Fig.  14  Effect of the number of hidden neurons on convergence time

    图  15  $\alpha$和$\beta$对收敛时间的影响

    Fig.  15  Influence of $\alpha$ and $\beta$ on convergence time

    图  16  ALRDBN训练结果

    Fig.  16  The training results of ALRDBN

    图  17  ALRDBN测试结果

    Fig.  17  The test results of ALRDBN

    图  18  ALRDBN训练RMSE

    Fig.  18  The training RMSE of ALRDBN

    图  19  隐含层神经元数对收敛时间的影响

    Fig.  19  Effect of the number of hidden neurons on convergence time

    图  20  $\alpha$和$\beta$对收敛时间的影响

    Fig.  20  Influence of $\alpha$ and $\beta$ on convergence time

    表  1  MNIST手写数字实验结果对比

    Table  1  Result comparison of MNIST experiment

    方法隐含层数每层节点数正确识别率运算时间(s)
    ALRDBN210093.1 %20.0
    CDBN210093.0 %34.3
    DBN[21]210092.6 %32.9
    下载: 导出CSV

    表  2  CO2浓度变化实验结果对比

    Table  2  Result comparison of CO2 forecasting

    方法网络结构RMSE (训练)RMSE (测试)运算时间(s)
    ALRDBN3-20-40-10.91641.16717.6
    DBN3-20-40-10.94871.283011.9
    CDBN[22]3-20-40-10.91331.150711.5
    BP3-60-1>0.11.3 ~ 6.615.8
    下载: 导出CSV

    表  3  Lorenz时序预测实验结果对比

    Table  3  Result comparison of Lorenz forecasting

    方法网络结构RMSE (训练)RMSE (测试)运算时间(s)
    ALRDBN3-3-3-0.02100.02252.9
    DBN3-3-3-10.03710.03883.6
    CDBN3-3-3-10.02080.02233.2
    BPNN[23]3-6-10.07000.0835>10
    SRNN[24]3-6-10.02320.03026.7
    下载: 导出CSV
  • [1] Bengio Y H, Delalleau O. On the expressive power of deep Architectures. In:Proceeding of the 22nd International Conference. Berlin Heidelberg, Germany:Springer-Verlag, 2011. 18-36
    [2] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks. Science, 2006, 313(5786):504-507 doi: 10.1126/science.1127647
    [3] 郭潇逍, 李程, 梅俏竹.深度学习在游戏中的应用.自动化学报, 2016, 42(5):676-684 http://www.aas.net.cn/CN/abstract/abstract18857.shtml

    Guo Xiao-Xiao, Li Cheng, Mei Qiao-Zhu. Deep learning applied to games. Acta Automatica Sinica, 2016, 42(5):676-684 http://www.aas.net.cn/CN/abstract/abstract18857.shtml
    [4] LeCun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553):436-444 doi: 10.1038/nature14539
    [5] 贺昱曜, 李宝奇.一种组合型的深度学习模型学习率策略.自动化学报, 2016, 42(6):953-958 http://www.aas.net.cn/CN/abstract/abstract18886.shtml

    He Yu-Yao, Li Bao-Qi. A combinatory form learning rate scheduling for deep learning model. Acta Automatica Sinica, 2016, 42(6):953-958 http://www.aas.net.cn/CN/abstract/abstract18886.shtml
    [6] 马帅, 沈韬, 王瑞琦, 赖华, 余正涛.基于深层信念网络的太赫兹光谱识别.光谱学与光谱分析, 2015, 35(12):3325-3329 http://cdmd.cnki.com.cn/Article/CDMD-10674-1015636153.htm

    Ma Shuai, Shen Tao, Wang Rui-Qi, Lai Hua, Yu Zheng-Tao. Terahertz spectroscopic identification with deep belief network. Spectroscopy and Spectral Analysis, 2015, 35(12):3325-3329 http://cdmd.cnki.com.cn/Article/CDMD-10674-1015636153.htm
    [7] 耿志强, 张怡康.一种基于胶质细胞链的改进深度信念网络模型.自动化学报, 2016, 42(6):943-952 http://www.aas.net.cn/CN/abstract/abstract18885.shtml

    Geng Zhi-Qiang, Zhang Yi-Kang. An improved deep belief network inspired by glia chains. Acta Automatica Sinica, 2016, 42(6):943-952 http://www.aas.net.cn/CN/abstract/abstract18885.shtml
    [8] Abdel-Zaher A M, Eldeib A M. Breast cancer classification using deep belief networks. Expert Systems with Applications, 2016, 46:139-144 doi: 10.1016/j.eswa.2015.10.015
    [9] Rumelhart D E, Hinton G E, Williams R J. Learning representations by back-propagating errors. Nature, 1986, 323(6088):533-536 doi: 10.1038/323533a0
    [10] Mohamed A R, Dahl G E, Hinton G. Acoustic modeling using deep belief networks. IEEE Transactions on Audio, Speech, and Language Processing, 2012, 20(1):14-22 doi: 10.1109/TASL.2011.2109382
    [11] Bengio Y. Learning deep architectures for AI. Foundations and Trends in Machine Learning, 2009, 2(1):1-127 doi: 10.1561/2200000006
    [12] Lopes N, Ribeiro B. Improving convergence of restricted Boltzmann machines via a learning adaptive step size. Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications, Lecture Notes in Computer Science. Berlin Heidelberg:Springer, 2012. 511-518
    [13] Raina R, Madhavan A, Ng A Y. Large-scale deep unsupervised learning using graphics processors. In:Proceedings of the 26th Annual International Conference on Machine Learning. New York, NY, USA:ACM, 2009. 873-880
    [14] Ly D L, Paprotski V, Danny Y. Neural Networks on GPUs:Restricted Boltzmann Machines, Technical Report, University of Toronto, Canada, 2009.
    [15] Lopes N, Ribeiro B. Towards adaptive learning with improved convergence of deep belief networks on graphics processing units. Pattern Recognition, 2014, 47(1):114-127 doi: 10.1016/j.patcog.2013.06.029
    [16] Le Roux N, Bengio Y. Representational power of restricted boltzmann machines and deep belief networks. Neural Computation, 2008, 20(6):1631-1649 doi: 10.1162/neco.2008.04-07-510
    [17] Hinton G E. Training products of experts by minimizing contrastive divergence. Neural Computation, 2002, 14(8):1771-1800 doi: 10.1162/089976602760128018
    [18] Yu X H, Chen G A, Cheng S X. Dynamic learning rate optimization of the backpropagation algorithm. IEEE Transactions on Neural Networks, 1995, 6(3):669-677 doi: 10.1109/72.377972
    [19] Magoulas G D, Vrahatis M N, Androulakis G S. Improving the convergence of the backpropagation algorithm using learning rate adaptation methods. Neural Computation, 1999, 11(7):1769-1796 doi: 10.1162/089976699300016223
    [20] Lee H, Ekanadham C, Ng A. Sparse deep belief net model for visual area V2. In:Proceedings of the 2008 Advances in Neural Information Processing Systems. Cambridge, MA:MIT Press, 2008. 873-880
    [21] Ji N N, Zhang J S, Zhang C X. A sparse-response deep belief network based on rate distortion theory. Pattern Recognition, 2014, 47(9):3179-3191 doi: 10.1016/j.patcog.2014.03.025
    [22] 乔俊飞, 潘广源, 韩红桂.一种连续型深度信念网的设计与应用.自动化学报, 2015, 41(12):2138-2146 http://www.aas.net.cn/CN/abstract/abstract18786.shtml

    Qiao Jun-Fei, Pan Guang-Yuan, Han Hong-Gui. Design and application of continuous deep belief network. Acta Automatica Sinica, 2015, 41(12):2138-2146 http://www.aas.net.cn/CN/abstract/abstract18786.shtml
    [23] Chang L C, Chen P A, Chang F J. Reinforced two-step-ahead weight adjustment technique for online training of recurrent neural networks. IEEE Transactions on Neural Networks and Learning Systems, 2012, 23(8):1269-1278 doi: 10.1109/TNNLS.2012.2200695
    [24] Chen Q L, Chai W, Qiao J F. A stable online self-constructing recurrent neural network. Advances in Neural Networks-ISNN 2011. Berlin Heidelberg:Springer, 2011, 6677:122-131
  • 加载中
图(20) / 表(3)
计量
  • 文章访问数:  2797
  • HTML全文浏览量:  373
  • PDF下载量:  1274
  • 被引次数: 0
出版历程
  • 收稿日期:  2016-05-10
  • 录用日期:  2016-10-09
  • 刊出日期:  2017-08-20

目录

    /

    返回文章
    返回