Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910207749128192 |
|---|---|
| author | Wang, Jinping Tong, Zixin Xie, Zhiwu Gao, Zhiqiang |
| author_facet | Wang, Jinping Tong, Zixin Xie, Zhiwu Gao, Zhiqiang |
| contents | 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 closer alignment with NC geometry while consistently outperforming strong long-tailed baselines on different datasets. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_10047 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View Wang, Jinping Tong, Zixin Xie, Zhiwu Gao, Zhiqiang Machine Learning Artificial Intelligence 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 closer alignment with NC geometry while consistently outperforming strong long-tailed baselines on different datasets. |
| title | Rethinking Loss Reweighting for Imbalance Learning as an Inverse Problem: A Neural Collapse Point of View |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2605.10047 |