Research map / Learning theory and algorithms / Prediction and decision losses
Label-Imbalanced and Group-Sensitive Classification under Overparameterization
Ganesh Ramachandra Kini, Orestis Paraskevas, Samet Oymak, Christos Thrampoulidis
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 |
|---|---|---|---|
| add multiplicative logit adjustment to LA-loss (VS vs LA) (ablation) | LA-loss · CIFAR-10 STEP-100 | top-1 accuracy (%): 78.23 -> 79.10 | Improved |
| add additive logit adjustment to CDT-loss (VS vs CDT) (ablation) | CDT-loss · CIFAR-10 STEP-100 | top-1 accuracy (%): 73.26 -> 79.10 | Improved |
| add multiplicative adjustment to LA-loss (VS vs LA) (ablation) | LA-loss · CIFAR-10 LT-100 | top-1 accuracy (%): 80.81 -> 80.82 | No clear effect |
| add multiplicative adjustment to LA-loss (VS vs LA) (ablation) | LA-loss · CIFAR-100 LT-100 | top-1 accuracy (%): 42.87 -> 43.52 | Improved |
| add multiplicative adjustment to LA-loss (VS vs LA) (ablation) | LA-loss · CIFAR-100 STEP-100 | top-1 accuracy (%): 45.69 -> 46.53 | Improved |
| group-sensitive VS-loss instead of CE | ResNet50 · Waterbirds | worst-group accuracy (%): 68.1 -> 76.7 | Improved |
| add DRO to Group VS (ablation) | Group VS · Waterbirds | worst-group accuracy (%): 76.7 -> 78.9 | Improved |
| use additive-only adjustments (Group LA) instead of CE | ResNet50 · Waterbirds | balanced accuracy (%): 84.9 -> 84.2 | No clear effect |
| add multiplicative adjustments (Group CDT instead of Group LA) (ablation) | Group LA · Waterbirds | worst-group accuracy (%): 70.1 -> 75.4 | Improved |
| add additive adjustments to Group CDT (Group VS vs Group CDT) (ablation) | Group CDT · Waterbirds | worst-group accuracy (%): 75.4 -> 76.7 | No clear effect |
| add additive adjustments to CDT with tau=-0.5 and train 300 epochs (ablation) | CDT-loss (gamma=0.15) · CIFAR-10 LT-100 | train accuracy at 200 epochs (%): ~93 -> ~97 | Improved |
| add additive adjustments to CDT when training to interpolation (300 epochs) (ablation) | CDT-loss · CIFAR-10 LT-100 | balanced test accuracy | No clear effect |
Closest papers by meaning
- Leveraging Group Classification with Descending Soft Labeling for Deep Imbalanced Regression (AAAI 2025 · similarity 0.90)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.89)
- Toward Robustness in Multi-Label Classification: A Data Augmentation Strategy against Imbalance and Noise (AAAI 2024 · similarity 0.87)
- The Group Robustness is in the Details: Revisiting Finetuning under Spurious Correlations (NeurIPS 2024 · similarity 0.87)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.87)
- Balanced Data, Imbalanced Spectra: Unveiling Class Disparities with Spectral Imbalance (ICML 2024 · similarity 0.87)
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.
