Learning with Multiple Complementary Labels

Lei Feng, Takuo Kaneko, Bo Han, Gang Niu, Bo An, Masashi Sugiyama
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:3072-3081, 2020.

Abstract

A complementary label (CL) simply indicates an incorrect class of an example, but learning with CLs results in multi-class classifiers that can predict the correct class. Unfortunately, the problem setting only allows a single CL for each example, which notably limits its potential since our labelers may easily identify multiple CLs (MCLs) to one example. In this paper, we propose a novel problem setting to allow MCLs for each example and two ways for learning with MCLs. In the first way, we design two wrappers that decompose MCLs into many single CLs, so that we could use any method for learning with CLs. However, the supervision information that MCLs hold is conceptually diluted after decomposition. Thus, in the second way, we derive an unbiased risk estimator; minimizing it processes each set of MCLs as a whole and possesses an estimation error bound. We further improve the second way into minimizing properly chosen upper bounds. Experiments show that the former way works well for learning with MCLs but the latter is even better.

Cite this Paper


BibTeX
@InProceedings{pmlr-v119-feng20a, title = {Learning with Multiple Complementary Labels}, author = {Feng, Lei and Kaneko, Takuo and Han, Bo and Niu, Gang and An, Bo and Sugiyama, Masashi}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {3072--3081}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/feng20a/feng20a.pdf}, url = {http://proceedings.mlr.press/v119/feng20a.html}, abstract = {A complementary label (CL) simply indicates an incorrect class of an example, but learning with CLs results in multi-class classifiers that can predict the correct class. Unfortunately, the problem setting only allows a single CL for each example, which notably limits its potential since our labelers may easily identify multiple CLs (MCLs) to one example. In this paper, we propose a novel problem setting to allow MCLs for each example and two ways for learning with MCLs. In the first way, we design two wrappers that decompose MCLs into many single CLs, so that we could use any method for learning with CLs. However, the supervision information that MCLs hold is conceptually diluted after decomposition. Thus, in the second way, we derive an unbiased risk estimator; minimizing it processes each set of MCLs as a whole and possesses an estimation error bound. We further improve the second way into minimizing properly chosen upper bounds. Experiments show that the former way works well for learning with MCLs but the latter is even better.} }
Endnote
%0 Conference Paper %T Learning with Multiple Complementary Labels %A Lei Feng %A Takuo Kaneko %A Bo Han %A Gang Niu %A Bo An %A Masashi Sugiyama %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-feng20a %I PMLR %P 3072--3081 %U http://proceedings.mlr.press/v119/feng20a.html %V 119 %X A complementary label (CL) simply indicates an incorrect class of an example, but learning with CLs results in multi-class classifiers that can predict the correct class. Unfortunately, the problem setting only allows a single CL for each example, which notably limits its potential since our labelers may easily identify multiple CLs (MCLs) to one example. In this paper, we propose a novel problem setting to allow MCLs for each example and two ways for learning with MCLs. In the first way, we design two wrappers that decompose MCLs into many single CLs, so that we could use any method for learning with CLs. However, the supervision information that MCLs hold is conceptually diluted after decomposition. Thus, in the second way, we derive an unbiased risk estimator; minimizing it processes each set of MCLs as a whole and possesses an estimation error bound. We further improve the second way into minimizing properly chosen upper bounds. Experiments show that the former way works well for learning with MCLs but the latter is even better.
APA
Feng, L., Kaneko, T., Han, B., Niu, G., An, B. & Sugiyama, M.. (2020). Learning with Multiple Complementary Labels. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:3072-3081 Available from http://proceedings.mlr.press/v119/feng20a.html.

Related Material