Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness Meets Uncertainty: Evidential Adversarial Training for Robust Selective Classification
Nicolas Sournac, Ahmed Baha Ben Jmaa, Bertrand Braeckeveldt
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 robust evidence alignment (beta>0) (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 20.11 -> 71.51 | Improved |
| add robust evidence alignment (beta>0) (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 37.67 -> 6.67 | Improved |
| add AWP (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 68.46 -> 71.51 | Improved |
| add AWP (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 8.21 -> 6.67 | Improved |
| use evidential clean loss instead of cross-entropy (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 69.56 -> 71.51 | Improved |
| use evidential clean loss instead of cross-entropy (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 7.88 -> 6.67 | Improved |
| align in log alpha instead of alpha (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 67.32 -> 71.51 | Improved |
| align in log alpha instead of Dirichlet mean (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 10.43 -> 6.67 | Improved |
| use IKL instead of KL for REA divergence (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 70.61 -> 71.51 | Improved |
| use IKL instead of L2 for REA divergence (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 7.64 -> 6.67 | Improved |
| use IKL instead of CE in inner maximization (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA AUGRC: 6.38 vs 6.67 | Mixed |
| use IKL instead of KL in inner maximization (ablation) | EV-AT · CIFAR-10 WRN-34-10 AugMix | Clean/AA accuracy: 71.52 vs 71.51 | No clear effect |
Closest papers by meaning
- Probabilistically Robust Learning: Balancing Average and Worst-case Performance (ICML 2022 · similarity 0.91)
- Robust Adversarial Quantification via Conflict-Aware Evidential Deep Learning (ICLR 2026 · similarity 0.90)
- Using Pre-Training Can Improve Model Robustness and Uncertainty (ICML 2019 · similarity 0.90)
- Adversarial Training for Probabilistic Robustness (ICCV 2025 · similarity 0.90)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.89)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.88)
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.
