Research map / Label-efficient and robust learning / Recognition and long-tail
Fine-Grained Class-Conditional Distribution Balancing for Debiased Learning
Miaoyun Zhao, Qiang Zhang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| replace single-Gaussian CCDB matching with fine-grained mode-wise matching (FG-CCDB) | CCDB · CelebA | WGA (%): 85.27 -> 89.22 | Improved |
| replace CCDB with FG-CCDB | CCDB · cMNIST 0.5% conflicting | accuracy (%): 83.20 -> 89.02 | Improved |
| replace CCDB with FG-CCDB | CCDB · Waterbirds | WGA (%): 90.48 -> 90.56 | No clear effect |
| replace CCDB with FG-CCDB | CCDB · cCIFAR10 0.5% conflicting | accuracy (%): 55.07 -> 55.28 | No clear effect |
| replace CCDB with FG-CCDB | CCDB · cCIFAR10 5% conflicting | accuracy (%): 74.64 -> 78.06 | Improved |
| use MST bias predictions instead of ground-truth annotations for GroupDRO (ablation) | GroupDRO · Waterbirds | WGA (%): 91.40 -> 88.47 | Hurt |
| use MST bias predictions instead of ground-truth annotations for DFR (ablation) | DFR · Waterbirds | WGA (%): 92.90 -> 91.49 | Hurt |
| use ground-truth annotations instead of MST for FG-CCDB (ablation) | FG-CCDB · Waterbirds and CelebA | WGA (%): Waterbirds 90.56 -> 91.76; CelebA 89.22 -> 89.09 | No clear effect |
| raise top-confidence ratio beta from 50% to 70% (ablation) | MST · smallest-mode F1 | F1: cCIFAR10 0.72 -> 0.79; CelebA 0.47 -> 0.40 | Mixed |
| use small initial-split fraction gamma (10%) instead of 50% (ablation) | MST · bias exploration | mode accuracy / smallest-mode recall | Improved |
| repeat bias enhancement learning (ablation) | MST · WGA across datasets | WGA | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SELF — SELF: Learning to Filter Noisy Labels with Self-Ensembling (ICLR 2020) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CDMAD: Class-Distribution-Mismatch-Aware Debiasing for Class-Imbalanced Semi-Supervised Learning (CVPR 2024 · similarity 0.86)
- Twice Class Bias Correction for Imbalanced Semi-supervised Learning (AAAI 2024 · similarity 0.86)
- LCGC: Learning from Consistency Gradient Conflicting for Class-Imbalanced Semi-Supervised Debiasing (AAAI 2025 · similarity 0.85)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.85)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.85)
- LT-Soups: Bridging Head and Tail Classes via Subsampled Model Soups (NeurIPS 2025 · similarity 0.85)
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.
