Research map / Label-efficient and robust learning / Recognition and long-tail
AutoBalance: Optimized Loss Functions for Imbalanced Data
Mingchen Li, Xuechen Zhang, Christos Thrampoulidis, Jiasi Chen, Samet Oymak
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 |
|---|---|---|---|
| Jointly tune l and Δ, initialized with LA loss, rather than use LA loss alone (ablation) | AutoBalance · CIFAR100-LT | balanced error: 58.96 -> 56.70 | Improved |
| Jointly tune l and Δ, initialized with LA loss, rather than use LA loss alone (ablation) | AutoBalance · ImageNet-LT | balanced error: 52.46 -> 50.91 | Improved |
| Jointly tune l and Δ, initialized with LA loss, rather than use LA loss alone (ablation) | AutoBalance · iNaturalist-2018 | balanced error: 34.06 -> 33.25 | Improved |
| Jointly tune l and Δ rather than tune l alone (ablation) | AutoBalance · CIFAR10-LT | balanced error: 23.02 -> 21.39 | Improved |
| Jointly tune l and Δ rather than tune l alone (ablation) | AutoBalance · CIFAR100-LT | balanced error: 58.71 -> 56.84 | Improved |
| Jointly tune l and Δ rather than tune Δ alone (ablation) | AutoBalance · ImageNet-LT | balanced error: 53.02 -> 51.74 | Improved |
| Jointly tune l and Δ rather than tune Δ alone (ablation) | AutoBalance · iNaturalist-2018 | balanced error: 34.37 -> 33.41 | Improved |
| Initialize joint l and Δ search with LA loss rather than balanced CE (ablation) | AutoBalance · ImageNet-LT | balanced error: 51.74 -> 50.91 | Improved |
| Use AutoBalance rather than the strongest reported baseline | AutoBalance · Modified Waterbird | DEO: 6.91 -> 4.25 | Improved |
| Use AutoBalance rather than the strongest reported baseline | AutoBalance · Modified Waterbird | balanced error: 16.47 -> 15.13 | Improved |
Closest papers by meaning
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.89)
- Automated Loss function Search for Class-imbalanced Node Classification (ICML 2024 · similarity 0.88)
- Learning under Localized Minority Imbalance (NeurIPS 2026 · similarity 0.85)
- Improved Balanced Classification with Theoretically Grounded Loss Functions (NeurIPS 2025 · similarity 0.85)
- Leveraging Group Classification with Descending Soft Labeling for Deep Imbalanced Regression (AAAI 2025 · similarity 0.85)
- Learning Imbalanced Data with Beneficial Label Noise (ICML 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.
