基于历史分类加权和分级竞争采样的多视角主动学习

doi:10.3969/j.issn.0372-2112.2017.01.007

摘要
图/表
参考文献(0)
相关文章 (15)

全文: PDF (1280 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要

多视角主动学习是一种相比于传统主动学习能够取得更大程度版本空间缩减的技术，已被应用于多种类型的大数据分析中.本文针对现有的多视角主动学习算法在分类假设生成和采样策略中存在的不足分别提出了相应的改进方案.本文将Boosting思想应用到多视角主动学习框架中，通过将历史上各次查询得到的分类假设进行加权式投票来实现每次查询后分类假设的强化；与此同时，还提出了一种自适应的分级竞争采样策略，当分类争议样本规模较大时通过无监督谱聚类获得上述样本的空间分布描述，并在各个聚类中结合样本的分类不确定度和冗余度信息通过二次规划求解以获得可靠的批处理采样.为了证明上述改进的有效性，本文将多视角主动学习应用到图像分类领域中，并通过基于不同图像特征的视角来分别生成相应的分类假设.实验表明，本文提出的两点改进策略不仅均有助于提升多视角主动学习的性能，而且基于上述不同视角随机组合的多视角主动学习方法相比于经典的单视角主动学习算法能够更快地实现收敛并达到较高的场景分类准确性.

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	姚拓中
	安鹏
	宋加涛

关键词 ：多视角主动学习, 分类器集成强化, 分级竞争采样, 图像分类

Abstract：

Multi-view active learning is a technique which can realize more significant reduction on version space than traditional active learning and has been used in large-scale data analysis.This paper proposes two improvements in both hypothesis generation and sampling strategy.We integrate boosting-like idea into the active learning framework which uses the weighted voting of all hypothetic outputs from the past queries.Furthermore,a novel adaptive hierarchical competition sampling is presented.In this sampling strategy,if the number of the contention samples is large,an unsupervised spectral clustering is activated to obtain the coarse distribution of these contention samples in the feature space and then both the classification uncertainty and redundancy measures are considered in each cluster to query the unlabeled samples in batch mode by solving quadratic programming.We apply multi-view active learning in image classification in order to prove the effectiveness of the improvements and different image features are used as views to generate the corresponding hypothesis.The experiments demonstrate that our two proposals can both efficiently improve the performance of the multi-view active learning and the random combination of these views can also obtain faster convergence and better classification accuracy than state-of-the-art single-view active learning algorithms.

Key words： multi-view active learning weighted hypothesis boosting hierarchical competition sampling image classification

收稿日期: 2015-05-20

ZTFLH:

TN911.73

基金资助:

浙江省自然科学基金（No.LY15F020011，No.LQ15F020004）；浙江省公益技术应用研究计划（No.2016C33255）；宁波市自然科学基金（No.2015A610132）

作者简介: 姚拓中,男.1983年2月出生,浙江宁波人.2011年毕业于浙江大学大学信电系,其后在中科院宁波工业技术研究院做博士后研究,2014年进入宁波工程学院电子与信息工程学院,主要从事计算机视觉和机器学习等方面的研究工作.E-mail:thomasyao@zju.edu.cn;安鹏,男,1981年11月出生,山西太原人,2009年毕业于清华大学工程物理系,同年就职于宁波工程学院电子与信息工程学院,主要从事图像处理和嵌入式系统设计的研究工作.E-mail:anp04@126.com

引用本文:

姚拓中, 安鹏, 宋加涛. 基于历史分类加权和分级竞争采样的多视角主动学习[J]. 电子学报, 2017, 45(1): 46-53. YAO Tuo-zhong, AN Peng, SONG Jia-tao. Multi-View Active Learning Based on Weighted Hypothesis Boosting and Hierarchical Competition Sampling. Acta Electronica Sinica, 2017, 45(1): 46-53.

链接本文:

http://www.ejournal.org.cn/CN/10.3969/j.issn.0372-2112.2017.01.007 或 http://www.ejournal.org.cn/CN/Y2017/V45/I1/46

[1] H A Simon,G Lea.Problem solving and rule education:a unified view knowledge and organization[J].Ernuam,1974,15(2):63-73.
[2] D Lewis,J Catlett.Heterogeneous uncertainty sampling for supervised learning[A].Proceedings of the 11th International Conference on Machine Learning[C].Rutgers University,New Brunswick,NJ,USA,1994.148-156.
[3] A McCallum,K Nigam.Employing EM in pool-based active learning for text classification[A].Proceedings of the 15th International Conference on Machine Learning[C].Madison,Wisconsin,USA,July 24-27,1998.359-367.
[4] D Cohn,L Atlas,R Ladner.Improving generalization with active learning[J].Machine Learning,1994,15(2):201-221.
[5] N Roy,A McCallum.Toward optimal active learning through sampling estimation of error reduction[A].Proceedings of the 18th International Conference on Machine Learning[C].Williamstown,MA,USA,2001.441-448.
[6] Y Freund,H S Seung,E Shamir,N Tishby.Selective sampling using the query by committee algorithm[J].Machine Learning,1997,28(2):133-168.
[7] N Abe,H Mamitsuka.Query learning strategies using boosting and bagging[A].Proceedings of the 15th International Conference on Machine Learning[C].Madison,Wisconsin,USA,July 24-27,1998.1-10.
[8] I Muslea,S Minton,C A Knoblock.Selective sampling with redundant views[C].Proceedings of the 17th National Conference on Artificial Intelligence,Austin,Texas,2000.621-626.
[9] A Blum,T Mitchell.Combining labeled and unlabeled data with co-training[A].Proceedings of the Workshop on Computational Learning Theory[C].Morgan Kaufmann,San Francisco,CA,1998.92-100.
[10] I Muslea,S Minton,C A Knoblock.Active+semi-supervised learning=robust multi-view learning[A].The 19th International Conference on Machine Learning[C].Sydney,Australia,July 8-12,2002.435-442.
[11] I Muslea,S Minton,C A Knoblock.Active learning with strong and weak views:a case study on wrapper induction[C].Proceedings of the International Joint Conference on Artificial Intelligence[C].Acapulco,Mexico,2003.415-420.
[12] C Dima,M Hebert.Active Learning for Outdoor Obstacle Detection[A].Proceedings of the Robotics Science and Systems[C].Massachusetts Institute of Technology,Cambridge,Massachusetts,2005.
[13] D Wei,M M Crawford.Active learning via multi-view and local proximity co-regularization for hyperspectral image classification[J].IEEE Journal of Selected Topics in Signal Processing,2011,5(3):618-628.
[14] J Cheng,K Q Wang.Active learning for image retrieval with Co-SVM[J].Pattern Recognition,2006,40(1):330-334.
[15] W Wang,Z H Zhou.On multi-view active learning and the combination with semi-supervised learning[A].International Conference on Machine Learning[C].Helsinki,Finland,2008.1152-1159.
[16] W Wang,Z H Zhou.Multi-view active learning in the non-realized case[A].Advances in Neural Information Processing Systems[C].Vancouver,Canada,2010.2388-2396.
[17] X Y Zhang,C S Xu,H Q Lu,S D Ma.Multi-view multi-label active learning for image classification[A].IEEE International Conference on Multimedia and Expo[C].New York,2009.258-261.
[18] W H Yang,G Q Liu,L Zhang,E H Chen.Multi-view learning with batch mode active selection for image retrieval[A].Proceedings of the 21st International Conference on Pattern Recognition[C].Tsukuba,2012.979-982.
[19] I Muslea,S Minton,C A Knoblock.Active learning with multiple views[J].Journal of Artificial Intelligence Research,2006:203-233.
[20] J B Shi,J Malik.Normalized cuts and image segmentation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2000,22(8):888-905.
[21] S J Huang,R Jin,Z H Zhou.Active learning by querying informative and representative examples[A].Advances in Neural Information Processing Systems[C].USA,2010.36(10):1936-1949.
[22] C K Dagli,S Rajaram,T S Huang.Leveraging active learning for relevance feedback using an information theoretic diversity measure[A].ACM Conference on Image and Video Retrieval[C].Lecture Notes in Computer Science,Tempe,AZ,USA,2006.123-132.
[23] S C H Hoi,R Jin,J K Zhu,M R Lyu.Semi-supervised SVM batch mode active learning for image retrieval[A].IEEE Conference on Computer Vision and Pattern Recognition[C].IEEE,2008.1-7.
[24] Fei-Fei Li,P Perona.A Bayesian hierarchical model for learning natural scene categories[A].IEEE Computer Society Conference on Computer Vision and Pattern Recognition[C].San Diego,CA,June 2005.524-531.
[25] S Tong,E Chang.Support vector machine active learning for image retrieval[A].Proceedings of 9th ACM Iinternational Cconference on Multimedia[C].New York,NY,USA,2001.107-118.