Research map / Learning theory and algorithms / Prediction and decision losses
Certifiably Adversarially Robust Detection of Out-of-Distribution Data
Julian Bitterwolf, Alexander Meinke, Matthias Hein
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 IBP-based worst-case confidence bound loss on OOD inputs (GOOD100 vs OE) | OOD detection training · MNIST FashionMNIST | guaranteed AUC (GAUC): 0.0 -> 78.0 | Improved |
| GOOD100 vs ACET on empirical adversarial AUC | OOD detection training · MNIST FashionMNIST | adversarial AUC (AAUC): 98.4 -> 96.5 | Hurt |
| use GOOD100 instead of plain training | MNIST classifier · MNIST | test accuracy: 99.4 -> 98.7 | Hurt |
| use GOOD100 instead of OE | SVHN classifier · SVHN | test accuracy: 95.5 -> 96.3 | Improved |
| use GOOD80 instead of OE | CIFAR-10 classifier · CIFAR-10 | test accuracy: 90.7 -> 90.1 | No clear effect |
| apply worst-case loss to all OOD samples (quantile 100) instead of 80 on CIFAR-10 (ablation) | GOOD · CIFAR-10 vs CIFAR-100 | clean AUC: 85.6 -> 70.0 | Hurt |
| apply worst-case loss to all OOD samples (quantile 100) instead of 80 on CIFAR-10 (ablation) | GOOD · CIFAR-10 vs CIFAR-100 | guaranteed AUC (GAUC): 42.3 -> 54.2 | Improved |
| add quantile restriction (GOOD80) instead of GOOD100 for the difficult CIFAR-10 task (ablation) | GOOD · CIFAR-10 vs CIFAR-100 | clean AUC: 70.0 -> 85.6 | Improved |
| add worst-case term for part of OOD samples (GOOD60 vs GOOD0) (ablation) | GOOD · CIFAR-10 vs CIFAR-100 | guaranteed AUC (GAUC): 0.0 -> 34.9 | Improved |
| GOOD100 vs OE on EMNIST letters detection | MNIST classifier · EMNIST Letters | AUC: 92.7 -> 99.0 | Improved |
| GOOD80 vs ACET on empirical AAUC on CIFAR-10 | CIFAR-10 OOD detection · CIFAR-10 vs CIFAR-100 | adversarial AUC (AAUC): 74.5 -> 48.2 | Hurt |
Closest papers by meaning
- Provably Adversarially Robust Detection of Out-of-Distribution Data (Almost) for Free (NeurIPS 2022 · similarity 0.92)
- Understanding Certified Training with Interval Bound Propagation (ICLR 2024 · similarity 0.86)
- Towards Robust Out-of-Distribution Generalization Bounds via Sharpness (ICLR 2024 · similarity 0.86)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.86)
- The Best of Both Worlds: On the Dilemma of Out-of-distribution Detection (NeurIPS 2024 · similarity 0.86)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · 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.
