Research map / Label-efficient and robust learning / Recognition and long-tail
Constrained Optimization to Train Neural Networks on Critical and Under-Represented Classes
Sara Sangalli, Ertunc Erdil, Andeas Hötker, Olivio F Donati, Ender Konukoglu
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 the asymmetric ALM constraint to BCE | Binary CIFAR10, imb. 100 · Binary CIFAR10, imb. 100 | Test AUC: 91.2 -> 93.1 | Improved |
| add the asymmetric ALM constraint to a baseline loss | Binary CIFAR10, imb. 100 · Binary CIFAR10, imb. 100 | FPR @ 98% TPR: 50.0 -> 45.0 | Improved |
| add the asymmetric ALM constraint to CB-BCE | Binary CIFAR100, imb. 100 · Binary CIFAR100, imb. 100 | Test AUC: 78.8 -> 83.8 | Improved |
| add the asymmetric ALM constraint to W-BCE | Binary CIFAR100, imb. 100 · Binary CIFAR100, imb. 100 | FPR @ 90% TPR: 41.0 -> 39.0 | Improved |
| add the asymmetric ALM constraint to BCE | Prostate MRI · Prostate MRI | AUC ens.: 70.9 -> 85.4 | Improved |
| add the second multi-class constraint, including a penalty on non-critical class probabilities (ablation) | ALM_m,1 + LDAM on long-tailed CIFAR10 · Long-tailed CIFAR10, imb. 100 | Overall Accuracy: 72.86 -> 73.32 | Improved |
| use the second multi-class constraint rather than the first (ablation) | CB-CE on long-tailed CIFAR100 · Long-tailed CIFAR100, imb. 1:20 | Overall Accuracy: 51.36 -> 50.73 | Hurt |
| directly optimize AUC with MBAUC rather than use the asymmetric ALM constraint | Binary CIFAR10, imb. 100 · Binary CIFAR10, imb. 100 | Test AUC: 74.0 vs 93.1 | Hurt |
| add the asymmetric ALM constraint to LDAM | Long-tailed CIFAR10, imb. 100 · Long-tailed CIFAR10, imb. 100 | Error @ 90% TPR: 29.86 -> 28.91 | Improved |
Closest papers by meaning
- Procrustean Training for Imbalanced Deep Learning (ICCV 2021 · similarity 0.85)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · similarity 0.85)
- Balancing the Scales: A Theoretical and Algorithmic Framework for Learning from Imbalanced Data (ICML 2025 · similarity 0.83)
- A Unified Loss for Handling Inter-Class and Intra-Class Imbalance in Medical Image Segmentation (AAAI 2025 · similarity 0.83)
- Restoring balance: principled under/oversampling of data for optimal classification (ICML 2024 · similarity 0.83)
- Improved Algorithm for Deep Active Learning under Imbalance via Optimal Separation (ICML 2025 · similarity 0.82)
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.
