Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification

Camille Garcin, Maximilien Servajean, Alexis Joly, Joseph Salmon
Proceedings of the 39th International Conference on Machine Learning, PMLR 162:7208-7222, 2022.

Abstract

In modern classification tasks, the number of labels is getting larger and larger, as is the size of the datasets encountered in practice. As the number of classes increases, class ambiguity and class imbalance become more and more problematic to achieve high top-1 accuracy. Meanwhile, Top-K metrics (metrics allowing K guesses) have become popular, especially for performance reporting. Yet, proposing top-K losses tailored for deep learning remains a challenge, both theoretically and practically. In this paper we introduce a stochastic top-K hinge loss inspired by recent developments on top-K calibrated losses. Our proposal is based on the smoothing of the top-K operator building on the flexible "perturbed optimizer" framework. We show that our loss function performs very well in the case of balanced datasets, while benefiting from a significantly lower computational time than the state-of-the-art top-K loss function. In addition, we propose a simple variant of our loss for the imbalanced case. Experiments on a heavy-tailed dataset show that our loss function significantly outperforms other baseline loss functions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v162-garcin22a, title = {Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification}, author = {Garcin, Camille and Servajean, Maximilien and Joly, Alexis and Salmon, Joseph}, booktitle = {Proceedings of the 39th International Conference on Machine Learning}, pages = {7208--7222}, year = {2022}, editor = {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan}, volume = {162}, series = {Proceedings of Machine Learning Research}, month = {17--23 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v162/garcin22a/garcin22a.pdf}, url = {https://proceedings.mlr.press/v162/garcin22a.html}, abstract = {In modern classification tasks, the number of labels is getting larger and larger, as is the size of the datasets encountered in practice. As the number of classes increases, class ambiguity and class imbalance become more and more problematic to achieve high top-1 accuracy. Meanwhile, Top-K metrics (metrics allowing K guesses) have become popular, especially for performance reporting. Yet, proposing top-K losses tailored for deep learning remains a challenge, both theoretically and practically. In this paper we introduce a stochastic top-K hinge loss inspired by recent developments on top-K calibrated losses. Our proposal is based on the smoothing of the top-K operator building on the flexible "perturbed optimizer" framework. We show that our loss function performs very well in the case of balanced datasets, while benefiting from a significantly lower computational time than the state-of-the-art top-K loss function. In addition, we propose a simple variant of our loss for the imbalanced case. Experiments on a heavy-tailed dataset show that our loss function significantly outperforms other baseline loss functions.} }
Endnote
%0 Conference Paper %T Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification %A Camille Garcin %A Maximilien Servajean %A Alexis Joly %A Joseph Salmon %B Proceedings of the 39th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2022 %E Kamalika Chaudhuri %E Stefanie Jegelka %E Le Song %E Csaba Szepesvari %E Gang Niu %E Sivan Sabato %F pmlr-v162-garcin22a %I PMLR %P 7208--7222 %U https://proceedings.mlr.press/v162/garcin22a.html %V 162 %X In modern classification tasks, the number of labels is getting larger and larger, as is the size of the datasets encountered in practice. As the number of classes increases, class ambiguity and class imbalance become more and more problematic to achieve high top-1 accuracy. Meanwhile, Top-K metrics (metrics allowing K guesses) have become popular, especially for performance reporting. Yet, proposing top-K losses tailored for deep learning remains a challenge, both theoretically and practically. In this paper we introduce a stochastic top-K hinge loss inspired by recent developments on top-K calibrated losses. Our proposal is based on the smoothing of the top-K operator building on the flexible "perturbed optimizer" framework. We show that our loss function performs very well in the case of balanced datasets, while benefiting from a significantly lower computational time than the state-of-the-art top-K loss function. In addition, we propose a simple variant of our loss for the imbalanced case. Experiments on a heavy-tailed dataset show that our loss function significantly outperforms other baseline loss functions.
APA
Garcin, C., Servajean, M., Joly, A. & Salmon, J.. (2022). Stochastic smoothing of the top-K calibrated hinge loss for deep imbalanced classification. Proceedings of the 39th International Conference on Machine Learning, in Proceedings of Machine Learning Research 162:7208-7222 Available from https://proceedings.mlr.press/v162/garcin22a.html.

Related Material