Metric Learning for Kernel Regression

Kilian Q. Weinberger, Gerald Tesauro
; Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics, PMLR 2:612-619, 2007.

Abstract

Kernel regression is a well-established method for nonlinear regression in which the target value for a test point is estimated using a weighted average of the surrounding training samples. The weights are typically obtained by applying a distance-based kernel function to each of the samples, which presumes the existence of a well-defined distance metric. In this paper, we construct a novel algorithm for supervised metric learning, which learns a distance function by directly minimizing the leave-one-out regression error. We show that our algorithm makes kernel regression comparable with the state of the art on several benchmark datasets, and we provide efficient implementation details enabling application to datasets with  O(10k) instances. Further, we show that our algorithm can be viewed as a supervised variation of PCA and can be used for dimensionality reduction and high dimensional data visualization.

Cite this Paper


BibTeX
@InProceedings{pmlr-v2-weinberger07a, title = {Metric Learning for Kernel Regression}, author = {Kilian Q. Weinberger and Gerald Tesauro}, booktitle = {Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics}, pages = {612--619}, year = {2007}, editor = {Marina Meila and Xiaotong Shen}, volume = {2}, series = {Proceedings of Machine Learning Research}, address = {San Juan, Puerto Rico}, month = {21--24 Mar}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v2/weinberger07a/weinberger07a.pdf}, url = {http://proceedings.mlr.press/v2/weinberger07a.html}, abstract = {Kernel regression is a well-established method for nonlinear regression in which the target value for a test point is estimated using a weighted average of the surrounding training samples. The weights are typically obtained by applying a distance-based kernel function to each of the samples, which presumes the existence of a well-defined distance metric. In this paper, we construct a novel algorithm for supervised metric learning, which learns a distance function by directly minimizing the leave-one-out regression error. We show that our algorithm makes kernel regression comparable with the state of the art on several benchmark datasets, and we provide efficient implementation details enabling application to datasets with  O(10k) instances. Further, we show that our algorithm can be viewed as a supervised variation of PCA and can be used for dimensionality reduction and high dimensional data visualization.} }
Endnote
%0 Conference Paper %T Metric Learning for Kernel Regression %A Kilian Q. Weinberger %A Gerald Tesauro %B Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2007 %E Marina Meila %E Xiaotong Shen %F pmlr-v2-weinberger07a %I PMLR %J Proceedings of Machine Learning Research %P 612--619 %U http://proceedings.mlr.press %V 2 %W PMLR %X Kernel regression is a well-established method for nonlinear regression in which the target value for a test point is estimated using a weighted average of the surrounding training samples. The weights are typically obtained by applying a distance-based kernel function to each of the samples, which presumes the existence of a well-defined distance metric. In this paper, we construct a novel algorithm for supervised metric learning, which learns a distance function by directly minimizing the leave-one-out regression error. We show that our algorithm makes kernel regression comparable with the state of the art on several benchmark datasets, and we provide efficient implementation details enabling application to datasets with  O(10k) instances. Further, we show that our algorithm can be viewed as a supervised variation of PCA and can be used for dimensionality reduction and high dimensional data visualization.
RIS
TY - CPAPER TI - Metric Learning for Kernel Regression AU - Kilian Q. Weinberger AU - Gerald Tesauro BT - Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics PY - 2007/03/11 DA - 2007/03/11 ED - Marina Meila ED - Xiaotong Shen ID - pmlr-v2-weinberger07a PB - PMLR SP - 612 DP - PMLR EP - 619 L1 - http://proceedings.mlr.press/v2/weinberger07a/weinberger07a.pdf UR - http://proceedings.mlr.press/v2/weinberger07a.html AB - Kernel regression is a well-established method for nonlinear regression in which the target value for a test point is estimated using a weighted average of the surrounding training samples. The weights are typically obtained by applying a distance-based kernel function to each of the samples, which presumes the existence of a well-defined distance metric. In this paper, we construct a novel algorithm for supervised metric learning, which learns a distance function by directly minimizing the leave-one-out regression error. We show that our algorithm makes kernel regression comparable with the state of the art on several benchmark datasets, and we provide efficient implementation details enabling application to datasets with  O(10k) instances. Further, we show that our algorithm can be viewed as a supervised variation of PCA and can be used for dimensionality reduction and high dimensional data visualization. ER -
APA
Weinberger, K.Q. & Tesauro, G.. (2007). Metric Learning for Kernel Regression. Proceedings of the Eleventh International Conference on Artificial Intelligence and Statistics, in PMLR 2:612-619

Related Material