Classification with Asymmetric Label Noise: Consistency and Maximal Denoising

Clayton Scott, Gilles Blanchard, Gregory Handy
; Proceedings of the 26th Annual Conference on Learning Theory, PMLR 30:489-511, 2013.

Abstract

In many real-world classification problems, the labels of training examples are randomly corrupted. Thus, the set of training examples for each class is contaminated by examples of the other class. Previous theoretical work on this problem assumes that the two classes are separable, that the label noise is independent of the true class label, or that the noise proportions for each class are known. We introduce a general framework for classification with label noise that eliminates these assumptions. Instead, we give assumptions ensuring identifiability and the existence of a consistent estimator of the optimal risk, with associated estimation strategies. For any arbitrary pair of contaminated distributions, there is a unique pair of non-contaminated distributions satisfying the proposed assumptions, and we argue that this solution corresponds in a certain sense to maximal denoising. In particular, we find that learning in the presence of label noise is possible even when the class-conditional distributions overlap and the label noise is not symmetric. A key to our approach is a universally consistent estimator of the maximal proportion of one distribution that is present in another, a problem we refer to as“mixture proportion estimation. This work is motivated by a problem in nuclear particle classification.

Cite this Paper


BibTeX
@InProceedings{pmlr-v30-Scott13, title = {Classification with Asymmetric Label Noise: Consistency and Maximal Denoising}, author = {Clayton Scott and Gilles Blanchard and Gregory Handy}, booktitle = {Proceedings of the 26th Annual Conference on Learning Theory}, pages = {489--511}, year = {2013}, editor = {Shai Shalev-Shwartz and Ingo Steinwart}, volume = {30}, series = {Proceedings of Machine Learning Research}, address = {Princeton, NJ, USA}, month = {12--14 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v30/Scott13.pdf}, url = {http://proceedings.mlr.press/v30/Scott13.html}, abstract = {In many real-world classification problems, the labels of training examples are randomly corrupted. Thus, the set of training examples for each class is contaminated by examples of the other class. Previous theoretical work on this problem assumes that the two classes are separable, that the label noise is independent of the true class label, or that the noise proportions for each class are known. We introduce a general framework for classification with label noise that eliminates these assumptions. Instead, we give assumptions ensuring identifiability and the existence of a consistent estimator of the optimal risk, with associated estimation strategies. For any arbitrary pair of contaminated distributions, there is a unique pair of non-contaminated distributions satisfying the proposed assumptions, and we argue that this solution corresponds in a certain sense to maximal denoising. In particular, we find that learning in the presence of label noise is possible even when the class-conditional distributions overlap and the label noise is not symmetric. A key to our approach is a universally consistent estimator of the maximal proportion of one distribution that is present in another, a problem we refer to as“mixture proportion estimation. This work is motivated by a problem in nuclear particle classification.} }
Endnote
%0 Conference Paper %T Classification with Asymmetric Label Noise: Consistency and Maximal Denoising %A Clayton Scott %A Gilles Blanchard %A Gregory Handy %B Proceedings of the 26th Annual Conference on Learning Theory %C Proceedings of Machine Learning Research %D 2013 %E Shai Shalev-Shwartz %E Ingo Steinwart %F pmlr-v30-Scott13 %I PMLR %J Proceedings of Machine Learning Research %P 489--511 %U http://proceedings.mlr.press %V 30 %W PMLR %X In many real-world classification problems, the labels of training examples are randomly corrupted. Thus, the set of training examples for each class is contaminated by examples of the other class. Previous theoretical work on this problem assumes that the two classes are separable, that the label noise is independent of the true class label, or that the noise proportions for each class are known. We introduce a general framework for classification with label noise that eliminates these assumptions. Instead, we give assumptions ensuring identifiability and the existence of a consistent estimator of the optimal risk, with associated estimation strategies. For any arbitrary pair of contaminated distributions, there is a unique pair of non-contaminated distributions satisfying the proposed assumptions, and we argue that this solution corresponds in a certain sense to maximal denoising. In particular, we find that learning in the presence of label noise is possible even when the class-conditional distributions overlap and the label noise is not symmetric. A key to our approach is a universally consistent estimator of the maximal proportion of one distribution that is present in another, a problem we refer to as“mixture proportion estimation. This work is motivated by a problem in nuclear particle classification.
RIS
TY - CPAPER TI - Classification with Asymmetric Label Noise: Consistency and Maximal Denoising AU - Clayton Scott AU - Gilles Blanchard AU - Gregory Handy BT - Proceedings of the 26th Annual Conference on Learning Theory PY - 2013/06/13 DA - 2013/06/13 ED - Shai Shalev-Shwartz ED - Ingo Steinwart ID - pmlr-v30-Scott13 PB - PMLR SP - 489 DP - PMLR EP - 511 L1 - http://proceedings.mlr.press/v30/Scott13.pdf UR - http://proceedings.mlr.press/v30/Scott13.html AB - In many real-world classification problems, the labels of training examples are randomly corrupted. Thus, the set of training examples for each class is contaminated by examples of the other class. Previous theoretical work on this problem assumes that the two classes are separable, that the label noise is independent of the true class label, or that the noise proportions for each class are known. We introduce a general framework for classification with label noise that eliminates these assumptions. Instead, we give assumptions ensuring identifiability and the existence of a consistent estimator of the optimal risk, with associated estimation strategies. For any arbitrary pair of contaminated distributions, there is a unique pair of non-contaminated distributions satisfying the proposed assumptions, and we argue that this solution corresponds in a certain sense to maximal denoising. In particular, we find that learning in the presence of label noise is possible even when the class-conditional distributions overlap and the label noise is not symmetric. A key to our approach is a universally consistent estimator of the maximal proportion of one distribution that is present in another, a problem we refer to as“mixture proportion estimation. This work is motivated by a problem in nuclear particle classification. ER -
APA
Scott, C., Blanchard, G. & Handy, G.. (2013). Classification with Asymmetric Label Noise: Consistency and Maximal Denoising. Proceedings of the 26th Annual Conference on Learning Theory, in PMLR 30:489-511

Related Material