Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Credal Ensembles via Distributionally Robust Optimization
Kaizheng Wang, Ghifari Adam Faza, Fabio Cuzzolin, Siu Lun Chau, David Moens, Hans Hallez
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 |
|---|---|---|---|
| use box credal set instead of convex hull (ablation) | CreDRO · CIFAR10 vs SVHN, M=5 | AUROC: 96.0 -> 96.6 | Improved |
| use box credal set instead of convex hull (ablation) | CreDRO · CIFAR10 vs SVHN, M=20 | AUROC: 97.2 -> 97.4 | Improved |
| increase ensemble size from 5 to 20 (ablation) | CreDRO · CIFAR10 vs SVHN | AUROC: 96.6 -> 97.4 | Improved |
| vary delta_G from 0.5 to 0.9 (ablation) | CreDRO · CIFAR10 vs SVHN | AUROC: 96.6 -> 96.1 | No clear effect |
| add top-loss selection instead of random upweighting at 10% label noise (ablation) | CreDRO · CIFAR10 vs SVHN, 10% noise | AUROC: 88.7 -> 92.2 | Improved |
| add top-loss selection instead of random upweighting at 20% label noise (ablation) | CreDRO · CIFAR10 vs SVHN, 20% noise | AUROC: 83.8 -> 88.9 | Improved |
| add DRO training to a classical ensemble (EN-DRO vs DE) | Deep ensemble · CIFAR10 vs SVHN, M=20 | AUROC: 94.8 -> 95.7 | Improved |
| add varied DRO levels and credal set (CreDRO vs EN-DRO) | EN-DRO · CIFAR10 vs SVHN, M=20 | AUROC: 95.7 -> 97.4 | Improved |
| use CreDRO averaged prediction instead of DE | Deep ensemble · CIFAR10 test | accuracy: 0.9569 -> 0.9637 | Improved |
| use CreDRO averaged prediction instead of DE | Deep ensemble · CIFAR10 test | ECE: 0.0051 -> 0.0038 | Improved |
| use DRO training instead of ERM (CreDRO vs CreWra) | CreWra · CIFAR10 training time | seconds: 6259.74 -> 6567.93 | Hurt |
Closest papers by meaning
- Credal Prediction based on Relative Likelihood (NeurIPS 2025 · similarity 0.92)
- Conformalized Credal Set Predictors (NeurIPS 2024 · similarity 0.89)
- Credal Learning Theory (NeurIPS 2024 · similarity 0.89)
- Credal Wrapper of Model Averaging for Uncertainty Estimation in Classification (ICLR 2025 · similarity 0.88)
- Credal Deep Ensembles for Uncertainty Quantification (NeurIPS 2024 · similarity 0.87)
- Rethinking Aleatoric and Epistemic Uncertainty (ICML 2025 · similarity 0.86)
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.
