Unsupervised Multiple Kernel Learning

Jinfeng Zhuang, Jialei Wang, Steven C. H. Hoi, Xiangyang Lan
Proceedings of the Asian Conference on Machine Learning, PMLR 20:129-144, 2011.

Abstract

Traditional multiple kernel learning (MKL) algorithms are essentially supervised learning in the sense that the kernel learning task requires the class labels of training data. However, class labels may not always be available prior to the kernel learning task in some real world scenarios, e.g., an early preprocessing step of a classification task or an unsupervised learning task such as dimension reduction. In this paper, we investigate a problem of Unsupervised Multiple Kernel Learning (UMKL), which does not require class labels of training data as needed in a conventional multiple kernel learning task. Since a kernel essentially defines pairwise similarity between any two examples, our unsupervised kernel learning method mainly follows two intuitive principles: (1) a good kernel should allow every example to be well reconstructed from its localized bases weighted by the kernel values; (2) a good kernel should induce kernel values that are coincided with the local geometry of the data. We formulate the unsupervised multiple kernel learning problem as an optimization task and propose an efficient alternating optimization algorithm to solve it. Empirical results on both classification and dimension reductions tasks validate the efficacy of the proposed UMKL algorithm.

Cite this Paper

BibTeX
@InProceedings{pmlr-v20-zhuang11, title = {Unsupervised Multiple Kernel Learning}, author = {Zhuang, Jinfeng and Wang, Jialei and Hoi, Steven C. H. and Lan, Xiangyang}, booktitle = {Proceedings of the Asian Conference on Machine Learning}, pages = {129--144}, year = {2011}, editor = {Hsu, Chun-Nan and Lee, Wee Sun}, volume = {20}, series = {Proceedings of Machine Learning Research}, address = {South Garden Hotels and Resorts, Taoyuan, Taiwain}, month = {14--15 Nov}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v20/zhuang11/zhuang11.pdf}, url = {https://proceedings.mlr.press/v20/zhuang11.html}, abstract = {Traditional multiple kernel learning (MKL) algorithms are essentially supervised learning in the sense that the kernel learning task requires the class labels of training data. However, class labels may not always be available prior to the kernel learning task in some real world scenarios, e.g., an early preprocessing step of a classification task or an unsupervised learning task such as dimension reduction. In this paper, we investigate a problem of Unsupervised Multiple Kernel Learning (UMKL), which does not require class labels of training data as needed in a conventional multiple kernel learning task. Since a kernel essentially defines pairwise similarity between any two examples, our unsupervised kernel learning method mainly follows two intuitive principles: (1) a good kernel should allow every example to be well reconstructed from its localized bases weighted by the kernel values; (2) a good kernel should induce kernel values that are coincided with the local geometry of the data. We formulate the unsupervised multiple kernel learning problem as an optimization task and propose an efficient alternating optimization algorithm to solve it. Empirical results on both classification and dimension reductions tasks validate the efficacy of the proposed UMKL algorithm.} }
Endnote
%0 Conference Paper %T Unsupervised Multiple Kernel Learning %A Jinfeng Zhuang %A Jialei Wang %A Steven C. H. Hoi %A Xiangyang Lan %B Proceedings of the Asian Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2011 %E Chun-Nan Hsu %E Wee Sun Lee %F pmlr-v20-zhuang11 %I PMLR %P 129--144 %U https://proceedings.mlr.press/v20/zhuang11.html %V 20 %X Traditional multiple kernel learning (MKL) algorithms are essentially supervised learning in the sense that the kernel learning task requires the class labels of training data. However, class labels may not always be available prior to the kernel learning task in some real world scenarios, e.g., an early preprocessing step of a classification task or an unsupervised learning task such as dimension reduction. In this paper, we investigate a problem of Unsupervised Multiple Kernel Learning (UMKL), which does not require class labels of training data as needed in a conventional multiple kernel learning task. Since a kernel essentially defines pairwise similarity between any two examples, our unsupervised kernel learning method mainly follows two intuitive principles: (1) a good kernel should allow every example to be well reconstructed from its localized bases weighted by the kernel values; (2) a good kernel should induce kernel values that are coincided with the local geometry of the data. We formulate the unsupervised multiple kernel learning problem as an optimization task and propose an efficient alternating optimization algorithm to solve it. Empirical results on both classification and dimension reductions tasks validate the efficacy of the proposed UMKL algorithm.
RIS
TY - CPAPER TI - Unsupervised Multiple Kernel Learning AU - Jinfeng Zhuang AU - Jialei Wang AU - Steven C. H. Hoi AU - Xiangyang Lan BT - Proceedings of the Asian Conference on Machine Learning DA - 2011/11/17 ED - Chun-Nan Hsu ED - Wee Sun Lee ID - pmlr-v20-zhuang11 PB - PMLR DP - Proceedings of Machine Learning Research VL - 20 SP - 129 EP - 144 L1 - http://proceedings.mlr.press/v20/zhuang11/zhuang11.pdf UR - https://proceedings.mlr.press/v20/zhuang11.html AB - Traditional multiple kernel learning (MKL) algorithms are essentially supervised learning in the sense that the kernel learning task requires the class labels of training data. However, class labels may not always be available prior to the kernel learning task in some real world scenarios, e.g., an early preprocessing step of a classification task or an unsupervised learning task such as dimension reduction. In this paper, we investigate a problem of Unsupervised Multiple Kernel Learning (UMKL), which does not require class labels of training data as needed in a conventional multiple kernel learning task. Since a kernel essentially defines pairwise similarity between any two examples, our unsupervised kernel learning method mainly follows two intuitive principles: (1) a good kernel should allow every example to be well reconstructed from its localized bases weighted by the kernel values; (2) a good kernel should induce kernel values that are coincided with the local geometry of the data. We formulate the unsupervised multiple kernel learning problem as an optimization task and propose an efficient alternating optimization algorithm to solve it. Empirical results on both classification and dimension reductions tasks validate the efficacy of the proposed UMKL algorithm. ER -
APA
Zhuang, J., Wang, J., Hoi, S.C.H. & Lan, X.. (2011). Unsupervised Multiple Kernel Learning. Proceedings of the Asian Conference on Machine Learning, in Proceedings of Machine Learning Research 20:129-144 Available from https://proceedings.mlr.press/v20/zhuang11.html.

Related Material