Confidence Scores Make Instance-dependent Label-noise Learning Possible

Antonin Berthon, Bo Han, Gang Niu, Tongliang Liu, Masashi Sugiyama
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:825-836, 2021.

Abstract

In learning with noisy labels, for every instance, its label can randomly walk to other classes following a transition distribution which is named a noise model. Well-studied noise models are all instance-independent, namely, the transition depends only on the original label but not the instance itself, and thus they are less practical in the wild. Fortunately, methods based on instance-dependent noise have been studied, but most of them have to rely on strong assumptions on the noise models. To alleviate this issue, we introduce confidence-scored instance-dependent noise (CSIDN), where each instance-label pair is equipped with a confidence score. We find that with the help of confidence scores, the transition distribution of each instance can be approximately estimated. Similarly to the powerful forward correction for instance-independent noise, we propose a novel instance-level forward correction for CSIDN. We demonstrate the utility and effectiveness of our method through multiple experiments on datasets with synthetic label noise and real-world unknown noise.

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-berthon21a, title = {Confidence Scores Make Instance-dependent Label-noise Learning Possible}, author = {Berthon, Antonin and Han, Bo and Niu, Gang and Liu, Tongliang and Sugiyama, Masashi}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {825--836}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/berthon21a/berthon21a.pdf}, url = {https://proceedings.mlr.press/v139/berthon21a.html}, abstract = {In learning with noisy labels, for every instance, its label can randomly walk to other classes following a transition distribution which is named a noise model. Well-studied noise models are all instance-independent, namely, the transition depends only on the original label but not the instance itself, and thus they are less practical in the wild. Fortunately, methods based on instance-dependent noise have been studied, but most of them have to rely on strong assumptions on the noise models. To alleviate this issue, we introduce confidence-scored instance-dependent noise (CSIDN), where each instance-label pair is equipped with a confidence score. We find that with the help of confidence scores, the transition distribution of each instance can be approximately estimated. Similarly to the powerful forward correction for instance-independent noise, we propose a novel instance-level forward correction for CSIDN. We demonstrate the utility and effectiveness of our method through multiple experiments on datasets with synthetic label noise and real-world unknown noise.} }
Endnote
%0 Conference Paper %T Confidence Scores Make Instance-dependent Label-noise Learning Possible %A Antonin Berthon %A Bo Han %A Gang Niu %A Tongliang Liu %A Masashi Sugiyama %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-berthon21a %I PMLR %P 825--836 %U https://proceedings.mlr.press/v139/berthon21a.html %V 139 %X In learning with noisy labels, for every instance, its label can randomly walk to other classes following a transition distribution which is named a noise model. Well-studied noise models are all instance-independent, namely, the transition depends only on the original label but not the instance itself, and thus they are less practical in the wild. Fortunately, methods based on instance-dependent noise have been studied, but most of them have to rely on strong assumptions on the noise models. To alleviate this issue, we introduce confidence-scored instance-dependent noise (CSIDN), where each instance-label pair is equipped with a confidence score. We find that with the help of confidence scores, the transition distribution of each instance can be approximately estimated. Similarly to the powerful forward correction for instance-independent noise, we propose a novel instance-level forward correction for CSIDN. We demonstrate the utility and effectiveness of our method through multiple experiments on datasets with synthetic label noise and real-world unknown noise.
APA
Berthon, A., Han, B., Niu, G., Liu, T. & Sugiyama, M.. (2021). Confidence Scores Make Instance-dependent Label-noise Learning Possible. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:825-836 Available from https://proceedings.mlr.press/v139/berthon21a.html.

Related Material