最大规范化依赖性多标记半监督学习方法

张晨光; 张燕; 张夏欢

doi:10.16383/j.aas.2015.c140893

最大规范化依赖性多标记半监督学习方法

doi: 10.16383/j.aas.2015.c140893

1.
海南大学信息科学技术学院海口 570228;
2.
北京凌云光视公司图像处理部北京 100097

基金项目:

国家自然科学基金(11261015),海南省高等学校科学研究项目(Hjkj2012-01)资助

详细信息

作者简介:
张晨光海南大学信息科学技术学院讲师.2009年获得北京工业大学硕士学位.主要研究方向为图像处理,模式识别.E-mail:huzcg@foxmail.com

张夏欢北京凌云光视公司图像处理部图像算法工程师.主要研究方向为图像处理和模式识别.E-mail:zhanggongzi@yahoo.cn

通讯作者:
张燕海南大学信息科学技术学院讲师.主要研究方向为数据分析和数据挖掘.本文通信作者.E-mail:zhangyanouc@sina.com

计量
- 文章访问数: 1787
- HTML全文浏览量: 108
- PDF下载量: 1120
- 被引次数: 0
出版历程
- 收稿日期: 2015-01-13
- 修回日期: 2015-05-06
- 刊出日期: 2015-09-20

Normalized Dependence Maximization Multi-label Semi-supervised Learning Method

1.
College of Information Science and Technology, Hainan University, Haikou 570228;
2.
Department of Image Processing, Luster LightTec, Beijing 100097

Funds:

Supported by National Natural Science Foundation of China (11261015), College Scientific Research Program of Hainan Province of China (Hjkj2012-01)

摘要

摘要: 针对现有多标记学习方法大多属于有监督学习方法, 而不能有效利用相对便宜且容易获得的大量未标记样本的问题, 本文提出了一种新的多标记半监督学习方法, 称为最大规范化依赖性多标记半监督学习方法(Normalized dependence maximization multi-label semi-supervised learning method). 该方法将已有标签作为约束条件,利用所有样本, 包括已标记和未标记样本,对特征集和标签集的规范化依赖性进行估计, 并以该估计值的最大化为目标, 最终通过求解带边界的迹比值问题为未标记样本打上标签. 与其他经典多标记学习方法在多个真实多标记数据集上的对比实验表明, 本文方法可以有效从已标记和未标记样本中学习, 尤其是已标记样本相对稀少时,学习效果得到了显著提高.
- 规范化依赖性 /
- 多标记学习 /
- 半监督学习 /
- 迹比值
Abstract: In view of the problems that most of present multi-label learning methods are supervised learning methods and cannot effectively make use of relatively inexpensive and easily obtained large number of unlabeled samples, this paper puts forward a new multi-label semi-supervised learning method, called normalized dependence maximization multi-label semi-supervised learning method (DMMS). The DMMS regards labeled samples as constraint conditions, estimates the normalized dependency of feature and label sets on all samples including labeled and unlabeled samples, and maximizes the estimation by finally addressing a trace ratio optimization problem with constraint conditions for label unlabeled samples. Experiments comparing DMMS with the state-of-the-art multi-label learning approaches on several real-world datasets show that the DMMS can effectively learn from labeled and unlabeled samples, especially when the labeled is relatively rare, the learning performance can be improved greatly.
- Normalized dependence /
- multi-label learning /
- semi-supervised learning /
- trace ratio

HTML全文

参考文献(29)

[1]	Tsoumakas G, Ioannis K, Ioannis V. Mining multi-label data. Data Mining and Knowledge Discovery Handbook. Berlin: Springer-Verlag, 2010. 667-685
[2]	Zhang M L, Zhou Z H. A review on multi-label learning algorithms. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(8): 1819-1837
[3]	Read J, Pfahringer B, Holmes G, Frank E. Classifier chains for multi-label classification. Machine Learning, 2011, 85(3): 333-359
[4]	Tsoumakas G, Katakis I, Vlahavas I. Random k-labelsets for multi-label classification. IEEE Transactions on Knowledge and Data Engineering, 2011, 23(7): 1079-1089
[5]	Tsoumakas G, Vlahavas I. Random k-labelsets: an ensemble method for multilabel classification. In: Proceedings of the 18th European Conference on Machine Learning. Warsaw, Poland: Springer, 2007. 406-417
[6]	Zhang M L, Zhou Z H. A k-nearest neighbor based algorithm for multi-label classification. In: Proceedings of the 2005 IEEE International Conference on Granular Computing. New York, USA: IEEE, 2005. 718-721
[7]	Elisseeff A, Weston J. A kernel method for multi-labelled classification. In: Proceedings of Advances in Neural Information Processing Systems. Cambridge, Massachusetts, USA: MIT Press, 2001. 681-687
[8]	Zha Z J, Mei T, Wang J D, Wang Z F, Hua X S. Graph-based semi-supervised learning with multiple labels. Journal of Visual Communication and Image Representation, 2009, 20(2): 97-103
[9]	Chen G, Song Y Q, Wang F, Zhang C S. Semi-supervised multi-label learning by solving a Sylvester equation. In: Proceedings of the 2008 SIAM International Conference on Data Mining. Atlanta, USA: Curran Associates, 2008. 410-419
[10]	Li Yu-Feng, Huang Sheng-Jun, Zhou Zhi-Hua. Regularized semi-supervised multi-label learning. Journal of Computer Research and Development, 2012, 49(6): 1272-1278(李宇峰, 黄圣君, 周志华. 一种基于正则化的半监督多标记学习方法. 计算机研究与发展, 2012, 49(6): 1272-1278)
[11]	Wang J D, Zhao Y H, Wu X Q, Hua X S. A transductive multi-label learning approach for video concept detection. Pattern Recognition, 2011, 44(10-11): 2274-2286
[12]	Guo Y H, Schuurmans D. Semi-supervised multi-label classification. In: Proceedings of the 2012 European Conference, Machine Learning and Knowledge Discovery in Databases. Bristol, UK: Springer, 2012. 355-370
[13]	Wu L, Zhang M L. Multi-Label classification with unlabeled data: an inductive approach. In: Proceedings of the 2013 Asian Conference on Machine Learning. Cambridge, Massachusetts, USA: MIT Press/JMLR, 2013. 197-212
[14]	Liu Y, Jin R, Yang L. Semi-supervised multi-label learning by constrained non-negative matrix factorization. In: Proceedings of the 21st National Conference on Artificial Intelligence. California, USA: AAAI Press, 2006. 421-426
[15]	Kong X N, Ng M K, Zhou Z H. Transductive multilabel learning via label set propagation. IEEE Transactions on Knowledge and Data Engineering, 2013, 25(3): 704-719
[16]	Gretton A, Smola A, Bousquet O, Herbrich R, Belitski A, Augath M, Murayama Y, Pauls J, Schölkopf B, Logothetis N. Kernel constrained covariance for dependence measurement. In: Proceedings of 10th International Workshop on Artificial Intelligence and Statistics. New Jersey, USA: Society for Artificial Intelligence and Statistics, 2005. 12-23
[17]	Gretton A, Bousquet B, Smola A, Schölkopf B. Measuring statistical dependence with Hilbert-Schmidt norms. In: Proceedings of 16th International Conference on Algorithmic Learning Theory. Singapore: Springer, 2005. 63-77
[18]	Bach F R, Jordan M I. Kernel independent component analysis. Journal of Machine Learning Research, 2002, 3: 1-48
[19]	Song L, Smola A, Gretton A, Borgwardt K M. A dependence maximization view of clustering. In: Proceedings of the 24th International Conference on Machine Learning. New York, USA: ACM, 2007. 815-822
[20]	Blaschko M, Gretton A. Learning taxonomies by dependence maximization. In: Proceedings of Advances in Neural Information Processing Systems. Cambridge, Massachusetts, USA: MIT Press, 2008. 153-160
[21]	Zhang Y, Zhou Z H. Multi-label dimensionality reduction via dependency maximization. In: Proceedings of the 23rd AAAI Conference on Artificial Intelligence. California, USA: AAAI Press, 2008. 1503-1505
[22]	Gretton A, Fukumizu K, Teo C H, L. Song, Schölkopf B, Smola A J. A kernel statistical test of independence. In: Proceedings of Advances in Neural Information Processing Systems. Cambridge, Massachusetts, USA: MIT Press, 2008. 582-592
[23]	Jia Y Q, Nie F P, Zhang C S. Trace ratio problem revisited. IEEE Transactions on Neural Networks, 2009, 20(4): 729-735
[24]	Wang H, Yan S C, Xu D, Tang X O, Huang T. Trace ratio vs. ratio trace for dimensionality reduction. In: Proceedings of the 2007 IEEE Conference on Computer Vision and Pattern Recognition. Minneapolis, MN: IEEE, 2007. 1-8
[25]	Boutell M R, Luo J B, Shen X P, Brown C M. Learning multi-label scene classification. Pattern Recognition, 2004, 37(9): 1757-1771
[26]	Trohidis K, Tsoumakas G, Kalliris G, Vlahavas I P. Multi-label classification of music into emotions. In: Proceedings of the 9th International Conference on Music Information Retrieval. Philadelphia, USA: Drexel University, 2008. 325-330
[27]	Katakis I, Tsoumakas G, Vlahavas I. Multilabel text classification for automated tag suggestion. In: Proceedings of the ECML/PKDD 2008 Discovery Challenge. Heidelberg, Berlin: Springer, 2008. 75-83
[28]	Lewis D D, Yang Y M, Rose T G, Li F. RCV1: a new benchmark collection for text categorization research. Journal of Machine Learning Research, 2004, 5: 361-397
[29]	Schapire R E, Singer Y. Boostexter: a boosting-based system for text categorization. Machine Learning, 2000, 39(2-3): 135-168