[1]侯 杰,茅耀斌,孙金生.一种最大化样本可分性半监督Boosting算法[J].南京理工大学学报(自然科学版),2014,38(05):675-681.
 Hou Jie,Mao Yaobin,Sun Jinsheng.Semi-supervised separability-maximum boosting[J].Journal of Nanjing University of Science and Technology,2014,38(05):675-681.
点击复制

一种最大化样本可分性半监督Boosting算法
分享到:

《南京理工大学学报》(自然科学版)[ISSN:1005-9830/CN:32-1397/N]

卷:
38卷
期数:
2014年05期
页码:
675-681
栏目:
出版日期:
2014-10-29

文章信息/Info

Title:
Semi-supervised separability-maximum boosting
作者:
侯 杰茅耀斌孙金生
南京理工大学 自动化学院,江苏 南京 210094
Author(s):
Hou JieMao YaobinSun Jinsheng
School of Automation,NUST,Nanjing 210094,China
关键词:
半监督学习 Boosting算法 可分性 分类器
Keywords:
semi-supervised learning boosting algorithm separability classifier
分类号:
TP181
摘要:
针对数据实际分布与假设不匹配时半监督学习算法难以改善分类器性能的问题,该文提出一种最大化样本可分性半监督Boosting算法,通过引入“高密度区域局部散度最小、样本空间全局散度最大”准则来学习未标注的样本。该准则使用两种半监督假设(聚类假设和流形假设),减少了因半监督假设与数据不匹配造成的准确率下降问题。实验结果表明,该文算法有效提高了Boosting算法在符合聚类假设数据集和符合流形假设数据集上的准确性,提高了分类器噪声数据的稳定性。
Abstract:
The semi-supervised learning often fails to improve classifier accuracy when the data distribution does not match its semi-supervised assumption.In this paper,the semi-supervised separability-maximum boosting(Semi-SMBoost),a novel semi-supervised boosting algorithm based on local minimizing in high-density region and global maximizing in sample space criterion is proposed.The criterion utilizes both the cluster assumption and the manifold assumption of the semi-supervised learning,and reduces the accuracy regression caused by miss matching of data distribution and semi-supervised assumption.Experiment results indicate that the Semi-SMBoost improves classification accuracy on both clustered data and manifold data on,and stabilizes the classifier on noised data.

参考文献/References:

[1] 余正涛,邹俊杰,赵兴,等.基于主动学习的最小二乘支持向量机稀疏化[J].南京理工大学学报,2012,36(1):12-17. Yu Zhengtao,Zou Junjie,Zhao Xing,et al.Sparseness of least squares support vector machines based on active learning[J].Journal of Nanjing University of Science and Technology,2012,36(1):12-17.
[2]崔鹏,张汝波.半监督系数选择法的人脸识别[J].哈尔滨工程大学学报,2012,33(7):855-861. Cui Peng,Zhang Rubo.A semi-supervised coefficient selection method for face recognition[J].Journal of Harbin Engineering University,2012,33(7):855-861.
[3]林亦宁,韦巍,戴渊明.基于双层粒子滤波和半监督Hough Forests的多目标跟踪[J].光电工程,2012,39(9):56-64. Lin Yiwei,Wei Wei,Dai Yuanming.Multi-objects tracking with dual-level particle filter embedded semi-supervised hough forests[J].Opto-electronic Engineering,2012,39(9):56-64.
[4]周旭东,陈晓红,陈松灿.半配对半监督场景下的低分辨率人脸识别[J].计算机研究与发展,2012,49(11):2328-2333. Zhou Xudong,Chen Xiaohong,Chen Songcan.Low-resolution face recognition in semi-paired and semi-supervised scenario[J].Journal of Computer Research and Development,2012,49(11):2328-2333.
[5]Chapelle O,Schölkopf B,Zien A.Semi-supervised learning[M].Cambridge,US:MIT Press,2006.
[6]Scudder III H.Probability of error of some adaptive pattern-recognition machines[J].Information Theory,IEEE Transactions on,1965,11(3):363-371.
[7]Blum A,Mitchell T.Combining labeled and unlabeled data with co-training[A].Proceedings of the Eleventh Annual Conference on Computational Learning Theory-COLT'98[C].New York,USA:ACM Press,1998,98(4):92-100.
[8]Joachims T.Transductive inference for text classification using support vector machines[A].ICML'99 Proceedings of the Sixteenth International Conference on Machine Learning[C].San Francisco,US:Morgan Kaufmann Publishers,1999:200-209.
[9]Bennett K P,Demiriz A,Maclin R.Exploiting unlabeled data in ensemble methods[A].Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining[C].Edmonton,Canada:ACM,2002:289-296.
[10]Blum A,Chawla S.Learning from labeled and unlabeled data using graph mincuts[A].Proceedings of 18th International Conference on Machine Learning[C].San Francisco,US:Morgan Kaufmann Publisher,2001:19-26.
[11]Joachims T.Transductive learning via spectral graph partitioning[A].Proceedings of the Twentieth International Conference on Machine Learning-ICML 2003[C].Washing DC,US:ICML,2003,20:290-297.
[12]Mallapragada P K,Jin Rong,Jain A K,et al.Semiboost:Boosting for semi-supervised learning[J].Pattern Analysis and Machine Intelligence,IEEE Transactions on,2009,31(11):2000-2014.
[13]Yang Jian,Zhang David,Yang Jingyu,et al.Globally maximizing,locally minimizing:unsupervised discriminant projection with applications to face and palm biometrics.[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2007,29(4):650-64.
[14]Zhu Xiaojin,Ghahramani Zoubin.Learning from labeled and unlabeled data with label propagation[A].Tech Rep CMUCALD02107[C].Pittsburgh,US:Center for Automated Learning and Discovery,CMU,2002,54(CMU-CALD-02-107):1-19.
[15]Goldberg A B,Zhu Xiaojin,Singh A,et al.Multi-manifold semi-supervised learning lives on multiple,intersecting manifolds[A].Proceedings of the Twelfth International Conference on Artificial Intelligence and Statistics(AISTATS2009)[C].Clearwater,US:AISTATS,2009:169-176.
[16]Grabner H,Leistner C,Bischof H.Semi-supervised on-line boosting for robust tracking[A].Proceedings of the 10th European Conference on Computer Vision[C].Marseille,France:Springer Berlin Heidelberg,2008:234-247.

备注/Memo

备注/Memo:
收稿日期:2013-01-23 修回日期:2013-03-29
基金项目:国家自然科学基金(60974129)
作者简介:侯杰(1985-),男,博士生,主要研究方向:人脸检测与跟踪、基于Boosting的特征选择,E-mail:reiase@gmail.com; 通讯作者:茅耀斌(1971-),男,博士,副教授,主要研究方向:图像处理与模式识别、多媒体信息安全,E-mail:maoybin@mail.njust.edu.cn。
引文格式:侯杰,茅耀斌,孙金生.一种最大化样本可分性半监督Boosting算法[J].南京理工大学学报,2014,38(5):675-681.
投稿网址:http://zrxuebao.njust.edu.cn
更新日期/Last Update: 2014-10-31