Research map / Adversarial robustness and security / Certified and empirical robustness
Evaluating Robustness of Predictive Uncertainty Estimation: Are Dirichlet-based Models Reliable?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply PGD label attacks with increasing radius | DBU models · CIFAR10 and Sensorless | AUC-PR for identifying correct vs. wrong predictions: 98.7 -> 0.4; 99.9 -> 3.2 | Hurt |
| Apply PGD label attacks to inputs | DBU models using uncertainty for attack detection · CIFAR10 and Sensorless | AUC-PR, label attack detection: CIFAR10 PostNet: 63.4 at radius 0.1; Sensorless PostNet: 85.0 at radius 2.0 | Mixed |
| Apply PGD uncertainty attacks to ID or OOD inputs | DBU models using differential entropy · CIFAR10–SVHN | AUC-PR, OOD detection: DDNet ID attack: 82.8 -> 15.4; OOD attack: 82.8 -> 15.4 | Hurt |
| Use median smoothing on uncertainty scores | DBU models · CIFAR10: correct-vs-wrong prediction, attack detection, and OOD detection | AUC-PR robustness | Helped |
| Add adversarial training using PGD label or uncertainty attacks | Smoothed DBU models · CIFAR10 robustness tasks | AUC-PR robustness | No clear effect |
| Train PriorNet with OOD data | PriorNet · Label attack robustness | Accuracy under label attacks | Helped |
Closest papers by meaning
- Discretization-Induced Dirichlet Posterior for Robust Uncertainty Quantification on Regression (AAAI 2024 · similarity 0.87)
- RobustBench: a standardized adversarial robustness benchmark (NeurIPS 2021 · similarity 0.87)
- Post-hoc Uncertainty Learning Using a Dirichlet Meta-Model (AAAI 2023 · similarity 0.86)
- ROME is Forged in Adversity: Robust Distilled Datasets via Information Bottleneck (ICML 2025 · similarity 0.85)
- OODRobustBench: a Benchmark and Large-Scale Analysis of Adversarial Robustness under Distribution Shift (ICML 2024 · similarity 0.85)
- Uncertainty Estimation by Flexible Evidential Deep Learning (NeurIPS 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.