·898· 智能系统学报 第12卷 明，尽管源任务与目标任务采用的数据集之间存在 chine learning research,2010,11:3371-3408. 差异，但是特征迁移模型仍然能够训练出目标数据 [11]CHEN M,XU Z,WINBERGER K Q,et al.Marginalized 集的图像特征，并且在目标任务的识别过程中能够 denoising autoencoders for domain adaptation[Cl//Interna- 达到优于经典深度学习模型的分类效果。 tional Conference on Machine Learning.Edinburgh,Scot- land,2012:1206-1214. 参考文献： [12]VINCENT P.A connection between score matching and [1]HINTON G E,ZEMEL R S.Autoencoder minimum de- denoising autoencoders[J].Neural computation,2011, scription length and helmholtz free energy[C]//Conference 23(7):1661-1674 on Neural Information Processing Systems(NIPS).Denver, [13]RIFAI S.Contractive auto-encoders:explicit invariance USA.1993:3-10 during feature extraction[Cl//Proceedings of the Twenty- [2]SOCHER R,HUANG E H,PENNINGTON J,et al.Dy- eight International Conference on Machine Learning.Bel- namic pooling and unfolding recursive autoencoders for levue,USA,2011:833-840 paraphrase detection[C]//Proc Neural Information and Pro- [14]RIFAI S,BENGIO Y,DAUPHIN Y,et al.A generative cessing Systems.Granada,Spain,2011:801-809. process for sampling contractive auto-encoders[C]//Proc [3]SWERSKY K,RANZATO M,BUCHMAN D,et al.On Int'l Conf Machine Learning.Edinburgh,Scotland,UK. score matching for energy based models:generalizing au- 2012:1206-1214. toencoders and simplifying deep learning[C]//Proc Int'l [15]LECUN Y.Neural networks:tricks of the trade(2nd ed.) Conf Machine Learning Bellevue.Washington,USA,2011: [M].Germany:Springer,2012:9-48 1201-1208. [16]ZEILER M D.TAYLOR G W.FERGUS R.Adaptive de- [4]FISCHER A,IGEL C.An introduction to restricted convolutional networks for mid and high level feature Boltzmann machines[Cl//Progress in Pattern Recognition. learning[C]//IEEE International Conference on Computer Image Analysis,Computer Vision,and Applications. Vision.Barcelona,Spain,2011:2013-2025. Guadalajara,Mexico,2012:14-36. [5]BREULEUX O,BENGIO Y,VINCENT P.Quickly gener- 作者简介： ating representative samples from an RBM-derived 杨梦铎，女，1989年生，讲师，博 process[J].Neural computation,2011,23(8):2053-2073. 土，主要研究方向为模式识别与机器 6]COURVILLE A,BERGSTRA J,BENGIO Y.Unsuper- 学习。 vised models of images by spike-and-slab RBMs[C]//Proc Int'l Conf Machine Learning.Bellevue,Washington,USA, 2011:1145-1152. [7]SCHMAH T,HINTON G E,ZEMEL R,et al.Generative versus discriminative training of RBMs for classification of 栾咏红，女，1968年生，副教授 fMRI images[C]//Proc Neural Information and Processing 主要研究方向为强化学习。 Systems.Vancouver,Canada.2008:1409-1416. [8]ERHAN D,BENGIO Y,COURVILLE A,et al.Why does unsupervised pre-training help deep learning?[J].Machine learning research,2010,11:625-660. [9]VINCENT P.Extracting and composing robust features 刘文军，男.1981年生，讲师，博 with denoising auto-encoders[C]//International Conference 士，主要研究方向为无线传感网络与 on Machine Learning(ICML).Helsinki,Finland,2008: 算法分析。 1096-1103 [10]VINCENT P,LAROCHELLE H,LAJOIE I,et al.Stacked denoising autoencoders:learning useful representations in a deep network with a local denoising criterion[J].Ma-明，尽管源任务与目标任务采用的数据集之间存在 差异，但是特征迁移模型仍然能够训练出目标数据 集的图像特征，并且在目标任务的识别过程中能够 达到优于经典深度学习模型的分类效果。 参考文献： HINTON G E, ZEMEL R S. Autoencoder minimum de￾scription length and helmholtz free energy[C]//Conference on Neural Information Processing Systems(NIPS). Denver, USA, 1993: 3–10. [1] SOCHER R, HUANG E H, PENNINGTON J, et al. Dy￾namic pooling and unfolding recursive autoencoders for paraphrase detection[C]//Proc Neural Information and Pro￾cessing Systems. Granada, Spain, 2011: 801–809. [2] SWERSKY K, RANZATO M, BUCHMAN D, et al. On score matching for energy based models: generalizing au￾toencoders and simplifying deep learning[C]//Proc Int’l Conf Machine Learning Bellevue. Washington, USA, 2011: 1201–1208. [3] FISCHER A, IGEL C. An introduction to restricted Boltzmann machines[C]//Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications. Guadalajara, Mexico, 2012: 14–36. [4] BREULEUX O, BENGIO Y, VINCENT P. Quickly gener￾ating representative samples from an RBM-derived process[J]. Neural computation, 2011, 23(8): 2053–2073. [5] COURVILLE A, BERGSTRA J, BENGIO Y. Unsuper￾vised models of images by spike-and-slab RBMs[C]//Proc Int’l Conf Machine Learning. Bellevue, Washington, USA, 2011: 1145–1152. [6] SCHMAH T, HINTON G E, ZEMEL R, et al. Generative versus discriminative training of RBMs for classification of fMRI images[C]//Proc Neural Information and Processing Systems. Vancouver, Canada, 2008: 1409–1416. [7] ERHAN D, BENGIO Y, COURVILLE A, et al. Why does unsupervised pre-training help deep learning?[J]. Machine learning research, 2010, 11: 625–660. [8] VINCENT P. Extracting and composing robust features with denoising auto-encoders[C]//International Conference on Machine Learning(ICML). Helsinki, Finland, 2008: 1096–1103. [9] VINCENT P, LAROCHELLE H, LAJOIE I, et al. Stacked denoising autoencoders: learning useful representations in a deep network with a local denoising criterion[J]. Ma- [10] chine learning research, 2010, 11: 3371–3408. CHEN M, XU Z, WINBERGER K Q, et al. Marginalized denoising autoencoders for domain adaptation[C]//Interna￾tional Conference on Machine Learning. Edinburgh, Scot￾land, 2012: 1206–1214. [11] VINCENT P. A connection between score matching and denoising autoencoders[J]. Neural computation, 2011, 23(7): 1661–1674. [12] RIFAI S. Contractive auto-encoders: explicit invariance during feature extraction[C]//Proceedings of the Twenty￾eight International Conference on Machine Learning. Bel￾levue, USA, 2011: 833–840 [13] RIFAI S, BENGIO Y, DAUPHIN Y, et al. A generative process for sampling contractive auto-encoders[C]//Proc Int’l Conf Machine Learning. Edinburgh, Scotland, UK, 2012: 1206–1214. [14] LECUN Y. Neural networks: tricks of the trade (2nd ed.) [M]. Germany: Springer, 2012: 9–48 [15] ZEILER M D, TAYLOR G W, FERGUS R. Adaptive de￾convolutional networks for mid and high level feature learning[C]//IEEE International Conference on Computer Vision. Barcelona, Spain, 2011: 2013–2025. [16] 作者简介： 杨梦铎，女，1989 年生，讲师，博 士，主要研究方向为模式识别与机器 学习。 栾咏红，女，1968 年生，副教授， 主要研究方向为强化学习。 刘文军，男，1981 年生，讲师，博 士，主要研究方向为无线传感网络与 算法分析。 ·898· 智 能 系 统 学 报 第 12 卷