AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

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

Jinping Wang, Zixin Tong, Zhiwu Xie, Zhiqiang Gao

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Classifier rebalancing and margin losses · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add batch-wise inverse reweighting (ablation)CE · CIFAR-100-LT IF=100Top-1 accuracy: 41.6 -> 46.7Improved
add macro-level reweighting (ablation)CE · CIFAR-100-LT IF=100Top-1 accuracy: 41.6 -> 45.4Improved
add macro-level reweighting on top of batch-wise inverse (ablation)CE · CIFAR-100-LT IF=100Top-1 accuracy: 46.7 -> 47.9Improved
replace cross-entropy by inverse-view reweightingResNet-32 · CIFAR-100-LT IF=100Top-1 accuracy: 41.62 -> 47.88Improved
replace inverse-frequency weighting by inverse-view reweightingResNet-32 · CIFAR-100-LT IF=200Top-1 accuracy: 25.25 -> 43.14Improved
add inverse-view loss reweightingGLMC · CIFAR-100-LT IF=100Top-1 accuracy: 55.9 -> 58.3Improved
add inverse-view loss reweightingCE* · CIFAR-10-LT IF=200Top-1 accuracy: 69.1 -> 78.6Improved
add learning-rate schedule on top of loss reweightingCE*+Loss · CIFAR-10-LT IF=100Top-1 accuracy: 82.3 -> 82.0No clear effect
add learning-rate schedule on top of loss reweightingCE*+Loss · CIFAR-100-LT IF=100Top-1 accuracy: 47.9 -> 48.9Improved
add learning-rate schedule on top of loss reweightingCE*+Loss · iNaturalist 2018Top-1 accuracy: 66.3 -> 65.8Hurt
add inverse-view loss reweightingCE* · ImageNet-LT FewTop-1 accuracy: 4.6 -> 8.3Improved
add inverse-view loss reweightingGLMC · ImageNet-LT AllTop-1 accuracy: 56.3 -> 57.6Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy