Classifier Cascade for Minimizing Feature Evaluation Cost

Minmin Chen, Zhixiang Xu, Kilian Weinberger, Olivier Chapelle, Dor Kedem
Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics, PMLR 22:218-226, 2012.

Abstract

Machine learning algorithms are increasingly used in large-scale industrial settings. Here, the operational cost during test-time has to be taken into account when an algorithm is designed. This operational cost is affected by the average running time and the computation time required for feature extraction. When a diverse set of features is used, the latter can vary drastically. In this paper we propose an algorithm that constructs a cascade of classifiers that explicitly trades-off operational cost and classifier accuracy while accounting for on-demand feature extraction costs. Different from previous work, our algorithm re-optimizes trained classifiers and allows expensive features to be scheduled at any stage within the cascade to minimize overall cost. Experiments on actual web-search ranking data sets demonstrate that our framework leads to drastic test-time improvements.

Cite this Paper


BibTeX
@InProceedings{pmlr-v22-chen12c, title = {Classifier Cascade for Minimizing Feature Evaluation Cost}, author = {Chen, Minmin and Xu, Zhixiang and Weinberger, Kilian and Chapelle, Olivier and Kedem, Dor}, booktitle = {Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics}, pages = {218--226}, year = {2012}, editor = {Lawrence, Neil D. and Girolami, Mark}, volume = {22}, series = {Proceedings of Machine Learning Research}, address = {La Palma, Canary Islands}, month = {21--23 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v22/chen12c/chen12c.pdf}, url = {https://proceedings.mlr.press/v22/chen12c.html}, abstract = {Machine learning algorithms are increasingly used in large-scale industrial settings. Here, the operational cost during test-time has to be taken into account when an algorithm is designed. This operational cost is affected by the average running time and the computation time required for feature extraction. When a diverse set of features is used, the latter can vary drastically. In this paper we propose an algorithm that constructs a cascade of classifiers that explicitly trades-off operational cost and classifier accuracy while accounting for on-demand feature extraction costs. Different from previous work, our algorithm re-optimizes trained classifiers and allows expensive features to be scheduled at any stage within the cascade to minimize overall cost. Experiments on actual web-search ranking data sets demonstrate that our framework leads to drastic test-time improvements.} }
Endnote
%0 Conference Paper %T Classifier Cascade for Minimizing Feature Evaluation Cost %A Minmin Chen %A Zhixiang Xu %A Kilian Weinberger %A Olivier Chapelle %A Dor Kedem %B Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2012 %E Neil D. Lawrence %E Mark Girolami %F pmlr-v22-chen12c %I PMLR %P 218--226 %U https://proceedings.mlr.press/v22/chen12c.html %V 22 %X Machine learning algorithms are increasingly used in large-scale industrial settings. Here, the operational cost during test-time has to be taken into account when an algorithm is designed. This operational cost is affected by the average running time and the computation time required for feature extraction. When a diverse set of features is used, the latter can vary drastically. In this paper we propose an algorithm that constructs a cascade of classifiers that explicitly trades-off operational cost and classifier accuracy while accounting for on-demand feature extraction costs. Different from previous work, our algorithm re-optimizes trained classifiers and allows expensive features to be scheduled at any stage within the cascade to minimize overall cost. Experiments on actual web-search ranking data sets demonstrate that our framework leads to drastic test-time improvements.
RIS
TY - CPAPER TI - Classifier Cascade for Minimizing Feature Evaluation Cost AU - Minmin Chen AU - Zhixiang Xu AU - Kilian Weinberger AU - Olivier Chapelle AU - Dor Kedem BT - Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics DA - 2012/03/21 ED - Neil D. Lawrence ED - Mark Girolami ID - pmlr-v22-chen12c PB - PMLR DP - Proceedings of Machine Learning Research VL - 22 SP - 218 EP - 226 L1 - http://proceedings.mlr.press/v22/chen12c/chen12c.pdf UR - https://proceedings.mlr.press/v22/chen12c.html AB - Machine learning algorithms are increasingly used in large-scale industrial settings. Here, the operational cost during test-time has to be taken into account when an algorithm is designed. This operational cost is affected by the average running time and the computation time required for feature extraction. When a diverse set of features is used, the latter can vary drastically. In this paper we propose an algorithm that constructs a cascade of classifiers that explicitly trades-off operational cost and classifier accuracy while accounting for on-demand feature extraction costs. Different from previous work, our algorithm re-optimizes trained classifiers and allows expensive features to be scheduled at any stage within the cascade to minimize overall cost. Experiments on actual web-search ranking data sets demonstrate that our framework leads to drastic test-time improvements. ER -
APA
Chen, M., Xu, Z., Weinberger, K., Chapelle, O. & Kedem, D.. (2012). Classifier Cascade for Minimizing Feature Evaluation Cost. Proceedings of the Fifteenth International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 22:218-226 Available from https://proceedings.mlr.press/v22/chen12c.html.

Related Material