伪特权信息和SVM+

doi:10.3969/j.issn.1001-2400.2016.06.018

西安电子科技大学学报

伪特权信息和SVM+

孙广玲;董勇;刘志

(上海大学通信与信息工程学院，上海 200072)

收稿日期:2015-10-26 出版日期:2016-12-20 发布日期:2017-01-19
作者简介:孙广玲(1975-)，女，副教授，E-mail: sunguangling@shu.edu.cn．
基金资助:
教育部科学技术研究重点资助项目(212053)；上海市自然科学基金资助项目(16ZR1411100)

Pseudo-privileged information and SVM+

SUN Guangling;DONG Yong;LIU Zhi

(School of Communication and Information Engineering, Shanghai Univ., Shanghai 200072, China)

Received:2015-10-26 Online:2016-12-20 Published:2017-01-19

摘要/Abstract

摘要：

针对只有部分训练样本拥有特权信息的特权学习，提出了伪特权信息及相应的SVM+．一方面，对于无特权信息的样本额外构造伪特权信息，使得这部分样本的松弛变量可在修正空间中预测，从而有效地提高了模型泛化能力．可用信息和随机特征都是有效的伪特权信息．另一方面，将真正特权信息用伪特权信息取代，使得全部训练样本的松弛变量都在惟一的修正空间中预测．在实践中发现，至少对于某些真正的特权信息和二分类问题来说，使用一个修正空间可获得更优的泛化能力．在手写数字和人脸表情识别问题上进行的实验结果显示，采用伪特权信息的SVM+具备一定的优势．

关键词: 特权学习, 伪特权信息, SVM+

Abstract:

In machine learning, learning using privileged information(LUPI) tries to improve the generalization of the classifier by leveraging information only available during learning. In the scenario of privileged information(PI) possessed by partial training samples, pseudo-privileged information(PPI) and SVM+are investigated. The proposed models depend on two formulations. One is to construct PPI for the samples without PI alone. The formulation enables slacks of such samples predicted in the correcting space with an ultimate goal of improving the generalization of the classifier. Available information and random features are proved to be effective options for PPI. The other is to replace the genuine PI with PPI so as to predict the slacks of all training samples in the unique correcting space. It is confirmed that at least for certain genuine PI and two categories classification task, the latter one is capable of obtaining better generalization performance. Experiments are performed on written digits and facial expression recognition. The results have validated advantages of SVM+using PPI.

Key words: learning using privileged information, pseudo-privileged information, SVM+

孙广玲;董勇;刘志. 伪特权信息和SVM+[J]. 西安电子科技大学学报, 2016, 43(6): 103-108.

SUN Guangling;DONG Yong;LIU Zhi. Pseudo-privileged information and SVM+[J]. Journal of Xidian University, 2016, 43(6): 103-108.

参考文献

［1］ VAPNIK V, VASHIST A, PAVLOVITCH N. Learning Using Hidden Information(Larning with Teacher)［C］//Proceedings of the International Joint Conference on Neural Networks. Piscataway: IEEE, 2009: 3188-3195.
［2］ VAPNIK V, VASHIST A. A New Learning Paradigm: Learning Using Privileged Information［J］. Neural Networks, 2009, 22(5/6): 544-557.
［3］ FEYEREISL J, AICKELIN U. Privileged Information for Data Clustering［J］. Information Sciences, 2012, 194: 4-23.
［4］ RIBEIRO B, SILVA C, VIEIRA A, et al. Financial Distress Model Prediction Using SVM+［C］//Proceedings of the International Joint Conference on Neural Networks. Piscataway: IEEE, 2010: 5596729.
［5］ PECHYONY D, VAPNIK V. On the Theory of Learnining with Privileged Information［C］//Advances in Neural Information Processing Systems 23: 24th Annual Conference on Neural Information Processing Systems. Vancouver: Neural Information Processing Systems Foundation, 2010: 1894-1902.
［6］ XU X X, LI W, XU D. Distance Metric Learning Using Privileged Information for Face Verification and Person Re-identification［J］. IEEE Transactions on Neural Networks and Learning Systems, 2015, 26(12): 3150-3162.
［7］ HERNNDEZ-LOBATO D, SHARMANSKA V, KERSTING K, et al. Mind the Nuisance: Gaussian Process Classification using Privileged Noise［C］//Advances in Neural Information Processing Systems 27: 28th Annual Conference on Neural Information Processing Systems.Vancouver: Neural Information Processing Systems Foundation, 2014: 837-845.
［8］ LAPIN M, HEIN M, SCHIELE B. Learning Using Privileged Information: SVM+and Weighted SVM［J］. Neural Networks, 2014, 53: 95-108.
［9］ FEYEREISL J, KWAK S, SON J, et al. Object Localization Based on Structural SVM Using Privileged Information［C］//Advances in Neural Information Processing Systems27: 28th Annual Conference on Neural Information Processing Systems. Vancouver: Neural Information Processing Systems Foundation, 2014: 208-216.
［10］ YANG H, PATRAS I. Privileged Information Based Conditional Structured Output Regression Forest for Facial Point Detection［J］. IEEE Transactions on Circuits and Systems for Video Technology, 2015, 25(9): 1507-1520.
［11］ SHARMANSKA V, QUADRIANTO N, LAMPERT C H. Learning to Rank Using Privileged Information［C］//Proceedings of the IEEE International Conference on Computer Vision. Piscataway: IEEE, 2013: 825-832.
［12］ WANG S Z, TAO D C, YANG J. Relative Attribute SVM+learning for Age Estimation［J］. IEEE Transactions on Cybernetics, 2016, 46(3): 827-839.
［13］ LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based Learning Applied to Document Recognition［J］. Proceedings of the IEEE, 1998, 86(11): 2278-2324.
［14］ SAVRAN A, ALYZ N, DIBEKLIOLU H, et al. Bosphorus Database for 3D Face Analysis［M］. Lecture Notes in Computer Science:5372. Heidelberg: Springer Verlag, 2008: 47-56.
［15］ AHONEN T, HADID A, PIETIKAINEN M. Face Description with Local Binary Patterns: Application to Face Recognition［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(12): 2037-2041.

伪特权信息和SVM+

Pseudo-privileged information and SVM+

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 0

Metrics

本文评价

推荐阅读 10