Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels

Yuxin Tian, Mouxing Yang, Yuhao Zhou, Jian Wang, Qing Ye, Tongliang Liu, Gang Niu, Jiancheng Lv
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:121042-121073, 2026.

Abstract

Conventional federated learning (FL) heavily depends on high-quality labels, which are often impractical in the real world, leading to the federated label-noise (F-LN) problem. Worse still, the F-LN problem is exacerbated by the heterogeneity of FL, whereas clients experience different label-noise types, ratios, and data distribution. In this study, we first observe an intriguing phenomenon that the global model of FL exhibits a slow memorization of noisy labels, suggesting its ability to maintain reliable predictions and robust representations in FL. Motivated by this, we propose a novel method termed Federated Global Reviser (FedGR), a straightforward yet effective method comprising three modules that collaboratively rectify noisy labels and regularize local training. By exploiting this inherent property, FedGR improves the label-noise robustness of FL in a self-contained manner. Extensive experiments on three widely used F-LN benchmarks demonstrate the superior performance of FedGR, consistently outperforming eight state-of-the-art baselines even in severe label-noise and data heterogeneity. Code: https://github.com/cs-yuxintian/FedGR-ICML26

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-tian26l, title = {Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels}, author = {Tian, Yuxin and Yang, Mouxing and Zhou, Yuhao and Wang, Jian and Ye, Qing and Liu, Tongliang and Niu, Gang and Lv, Jiancheng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {121042--121073}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/tian26l/tian26l.pdf}, url = {https://proceedings.mlr.press/v306/tian26l.html}, abstract = {Conventional federated learning (FL) heavily depends on high-quality labels, which are often impractical in the real world, leading to the federated label-noise (F-LN) problem. Worse still, the F-LN problem is exacerbated by the heterogeneity of FL, whereas clients experience different label-noise types, ratios, and data distribution. In this study, we first observe an intriguing phenomenon that the global model of FL exhibits a slow memorization of noisy labels, suggesting its ability to maintain reliable predictions and robust representations in FL. Motivated by this, we propose a novel method termed Federated Global Reviser (FedGR), a straightforward yet effective method comprising three modules that collaboratively rectify noisy labels and regularize local training. By exploiting this inherent property, FedGR improves the label-noise robustness of FL in a self-contained manner. Extensive experiments on three widely used F-LN benchmarks demonstrate the superior performance of FedGR, consistently outperforming eight state-of-the-art baselines even in severe label-noise and data heterogeneity. Code: https://github.com/cs-yuxintian/FedGR-ICML26} }
Endnote
%0 Conference Paper %T Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels %A Yuxin Tian %A Mouxing Yang %A Yuhao Zhou %A Jian Wang %A Qing Ye %A Tongliang Liu %A Gang Niu %A Jiancheng Lv %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-tian26l %I PMLR %P 121042--121073 %U https://proceedings.mlr.press/v306/tian26l.html %V 306 %X Conventional federated learning (FL) heavily depends on high-quality labels, which are often impractical in the real world, leading to the federated label-noise (F-LN) problem. Worse still, the F-LN problem is exacerbated by the heterogeneity of FL, whereas clients experience different label-noise types, ratios, and data distribution. In this study, we first observe an intriguing phenomenon that the global model of FL exhibits a slow memorization of noisy labels, suggesting its ability to maintain reliable predictions and robust representations in FL. Motivated by this, we propose a novel method termed Federated Global Reviser (FedGR), a straightforward yet effective method comprising three modules that collaboratively rectify noisy labels and regularize local training. By exploiting this inherent property, FedGR improves the label-noise robustness of FL in a self-contained manner. Extensive experiments on three widely used F-LN benchmarks demonstrate the superior performance of FedGR, consistently outperforming eight state-of-the-art baselines even in severe label-noise and data heterogeneity. Code: https://github.com/cs-yuxintian/FedGR-ICML26
APA
Tian, Y., Yang, M., Zhou, Y., Wang, J., Ye, Q., Liu, T., Niu, G. & Lv, J.. (2026). Learning Locally, Revising Globally: Global Reviser for Federated Learning with Noisy Labels. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:121042-121073 Available from https://proceedings.mlr.press/v306/tian26l.html.

Related Material