Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View

Jinping Wang, Zixin Tong, Zhiwu Xie, Zhiqiang Gao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:130142-130161, 2026.

Abstract

Loss reweighting is a widely used strategy for long-tailed classification, but existing reweighting strategies often rely on heuristics and rarely define a well-specified target. Inspired by Neural Collapse (NC), the ideal simplex Equiangular Tight Frame (ETF) terminal geometry suggests equal per-class average loss as a reasonable target for reweighting. Based on the ideal equal loss objective, we consider loss reweighting as an inverse problem and propose an inverse-view reweighting strategy that infers class weights dynamically to match this ideal objective. Empirically, NC metrics suggest our method can effectively reduce the loss imbalance coefficient and achieve closer alignment with NC geometry while consistently outperforming strong long-tailed baselines on different datasets. Our code is publicly available at: https://github.com/tongzixin716716/Inverse-Loss-Reweighting.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-wang26hj, title = {Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View}, author = {Wang, Jinping and Tong, Zixin and Xie, Zhiwu and Gao, Zhiqiang}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {130142--130161}, 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/wang26hj/wang26hj.pdf}, url = {https://proceedings.mlr.press/v306/wang26hj.html}, abstract = {Loss reweighting is a widely used strategy for long-tailed classification, but existing reweighting strategies often rely on heuristics and rarely define a well-specified target. Inspired by Neural Collapse (NC), the ideal simplex Equiangular Tight Frame (ETF) terminal geometry suggests equal per-class average loss as a reasonable target for reweighting. Based on the ideal equal loss objective, we consider loss reweighting as an inverse problem and propose an inverse-view reweighting strategy that infers class weights dynamically to match this ideal objective. Empirically, NC metrics suggest our method can effectively reduce the loss imbalance coefficient and achieve closer alignment with NC geometry while consistently outperforming strong long-tailed baselines on different datasets. Our code is publicly available at: https://github.com/tongzixin716716/Inverse-Loss-Reweighting.} }
Endnote
%0 Conference Paper %T Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View %A Jinping Wang %A Zixin Tong %A Zhiwu Xie %A Zhiqiang Gao %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-wang26hj %I PMLR %P 130142--130161 %U https://proceedings.mlr.press/v306/wang26hj.html %V 306 %X Loss reweighting is a widely used strategy for long-tailed classification, but existing reweighting strategies often rely on heuristics and rarely define a well-specified target. Inspired by Neural Collapse (NC), the ideal simplex Equiangular Tight Frame (ETF) terminal geometry suggests equal per-class average loss as a reasonable target for reweighting. Based on the ideal equal loss objective, we consider loss reweighting as an inverse problem and propose an inverse-view reweighting strategy that infers class weights dynamically to match this ideal objective. Empirically, NC metrics suggest our method can effectively reduce the loss imbalance coefficient and achieve closer alignment with NC geometry while consistently outperforming strong long-tailed baselines on different datasets. Our code is publicly available at: https://github.com/tongzixin716716/Inverse-Loss-Reweighting.
APA
Wang, J., Tong, Z., Xie, Z. & Gao, Z.. (2026). Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:130142-130161 Available from https://proceedings.mlr.press/v306/wang26hj.html.

Related Material