Research map / Label-efficient and robust learning / Recognition and long-tail
Better May Not Be Fairer: A Study on Subgroup Discrepancy in Image Classification
Ming-Chang Chiu, Pin-Yu Chen, Xuezhe Ma
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 |
|---|---|---|---|
| apply FlowAug flow-latent augmentation instead of Standard training | ResNet18 · CIFAR-10 | accuracy, best: 95.16 -> 96.58 | Improved |
| apply FlowAug instead of Standard training | ResNet18 · CIFAR-100 | accuracy, best: 78.52 -> 79.99 | Improved |
| apply FlowAug instead of Standard training | ResNet18 · CIFAR10-B | MacroStd: 2.24 -> 1.65 | Improved |
| perturb or mix the local representation nu instead of global z (ablation) | FlowAug · CIFAR-10 / CIFAR-100 | accuracy, best: CIFAR-10 96.58 -> 94.22; CIFAR-100 79.99 -> 74.05 | Hurt |
| use uniform noise instead of truncated Gaussian for z perturbation (ablation) | FlowAug · CIFAR-10 | accuracy, best: 96.23 -> 96.12 | No clear effect |
| combine FlowAug with Cutmix (ablation) | FlowAug · CIFAR-100 | accuracy: 79.45 -> 81.57 | Improved |
| add Standard loss and Mix z to FlowAug with uniform noise (ablation) | FlowAug · CIFAR-10 | accuracy, best: 96.12 -> 96.58 | Improved |
| apply Mixup and Cutmix on CIFAR-100 subgroup discrepancy | ResNet18 · CIFAR100-B | MacroStd: 12.24 -> 12.62 (Cutmix) | Hurt |
| use MacroStd instead of WeightedStd as sensitivity metric | metric · CIFAR-10 | correlation with accuracy: -0.78 -> -0.89 | Improved |
| apply FlowAug on ImageNet-10 | ResNet18 · ImageNet-10 | accuracy: 91.80 -> 94.80 | Improved |
Closest papers by meaning
- Discover and Mitigate Multiple Biased Subgroups in Image Classifiers (CVPR 2024 · similarity 0.90)
- Identification of Systematic Errors of Image Classifiers on Rare Subgroups (ICCV 2023 · similarity 0.88)
- Subgroups Matter for Robust Bias Mitigation (ICML 2025 · similarity 0.86)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.85)
- Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotations (CVPR 2024 · similarity 0.85)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.84)
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.
