CrossSplit: Mitigating Label Noise Memorization through Data Splitting

Jihye Kim, Aristide Baratin, Yan Zhang, Simon Lacoste-Julien
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:16377-16392, 2023.

Abstract

We approach the problem of improving robustness of deep learning algorithms in the presence of label noise. Building upon existing label correction and co-teaching methods, we propose a novel training procedure to mitigate the memorization of noisy labels, called CrossSplit, which uses a pair of neural networks trained on two disjoint parts of the labeled dataset. CrossSplit combines two main ingredients: (i) Cross-split label correction. The idea is that, since the model trained on one part of the data cannot memorize example-label pairs from the other part, the training labels presented to each network can be smoothly adjusted by using the predictions of its peer network; (ii) Cross-split semi-supervised training. A network trained on one part of the data also uses the unlabeled inputs of the other part. Extensive experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet and mini-WebVision datasets demonstrate that our method can outperform the current state-of-the-art in a wide range of noise ratios. The project page is at https://rlawlgul.github.io/.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-kim23a, title = {{C}ross{S}plit: Mitigating Label Noise Memorization through Data Splitting}, author = {Kim, Jihye and Baratin, Aristide and Zhang, Yan and Lacoste-Julien, Simon}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {16377--16392}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/kim23a/kim23a.pdf}, url = {https://proceedings.mlr.press/v202/kim23a.html}, abstract = {We approach the problem of improving robustness of deep learning algorithms in the presence of label noise. Building upon existing label correction and co-teaching methods, we propose a novel training procedure to mitigate the memorization of noisy labels, called CrossSplit, which uses a pair of neural networks trained on two disjoint parts of the labeled dataset. CrossSplit combines two main ingredients: (i) Cross-split label correction. The idea is that, since the model trained on one part of the data cannot memorize example-label pairs from the other part, the training labels presented to each network can be smoothly adjusted by using the predictions of its peer network; (ii) Cross-split semi-supervised training. A network trained on one part of the data also uses the unlabeled inputs of the other part. Extensive experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet and mini-WebVision datasets demonstrate that our method can outperform the current state-of-the-art in a wide range of noise ratios. The project page is at https://rlawlgul.github.io/.} }
Endnote
%0 Conference Paper %T CrossSplit: Mitigating Label Noise Memorization through Data Splitting %A Jihye Kim %A Aristide Baratin %A Yan Zhang %A Simon Lacoste-Julien %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-kim23a %I PMLR %P 16377--16392 %U https://proceedings.mlr.press/v202/kim23a.html %V 202 %X We approach the problem of improving robustness of deep learning algorithms in the presence of label noise. Building upon existing label correction and co-teaching methods, we propose a novel training procedure to mitigate the memorization of noisy labels, called CrossSplit, which uses a pair of neural networks trained on two disjoint parts of the labeled dataset. CrossSplit combines two main ingredients: (i) Cross-split label correction. The idea is that, since the model trained on one part of the data cannot memorize example-label pairs from the other part, the training labels presented to each network can be smoothly adjusted by using the predictions of its peer network; (ii) Cross-split semi-supervised training. A network trained on one part of the data also uses the unlabeled inputs of the other part. Extensive experiments on CIFAR-10, CIFAR-100, Tiny-ImageNet and mini-WebVision datasets demonstrate that our method can outperform the current state-of-the-art in a wide range of noise ratios. The project page is at https://rlawlgul.github.io/.
APA
Kim, J., Baratin, A., Zhang, Y. & Lacoste-Julien, S.. (2023). CrossSplit: Mitigating Label Noise Memorization through Data Splitting. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:16377-16392 Available from https://proceedings.mlr.press/v202/kim23a.html.

Related Material