AI Paper Atlas

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

Class-Conditional Distribution Balancing for Group Robust Classification

Miaoyun Zhao, Qiang Zhang

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 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
use class-conditional distribution balancing instead of class-balancing-only reweighting (ReWeight, BSoftmax, ReWeightCRT) (ablation)CCDB · Waterbirds worst-groupworst-group accuracy (%): 76.30 -> 90.48 (vs ReWeightCRT)Improved
train biased feature extractor on a smaller data split (gamma 30-50%) instead of full data (ablation)CCDB stage 1 · cMNIST / Waterbirdsworst-group accuracyImproved
reduce gamma to 10% (ablation)CCDB stage 1 · Waterbirdsspurious dominance of featuresHurt
clip sample-weight logits to [-T, T] (ablation)CCDB stage 2 · Waterbirds / cCIFAR10overall performanceImproved
CCDB vs ERM on cMNIST with 0.5% bias-conflicting samplesERM · cMNIST (0.5%)test accuracy (%): 35.19 -> 83.20Improved
CCDB vs ERM on CelebAERM · CelebAworst-group accuracy (%): 47.20 -> 85.27Improved
CCDB vs ERM on WaterbirdsERM · Waterbirdsworst-group accuracy (%): 72.60 -> 90.48Improved
CCDB vs ERM on CivilCommentsERM · CivilCommentsworst-group accuracy (%): 66.7 -> 75.00Improved
CCDB vs bias-supervised GroupDRO on CelebAGroupDRO · CelebAworst-group accuracy (%): 88.90 -> 85.27Hurt
CCDB vs bias-supervised GroupDRO on cCIFAR10 (multi-class)GroupDRO · cCIFAR10 (0.5%)test accuracy (%): 33.44 -> 55.07Improved
apply sample reweighting to features (ablation)CCDB · feature correlations with bias vs labelsfeature spuriousnessImproved

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