Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness of Bayesian Neural Networks to Gradient-Based Attacks
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 |
|---|---|---|---|
| use gradient-based attack (FGSM/PGD) instead of random sign attack | BNN (HMC, VI) · MNIST and Fashion MNIST | adversarial accuracy: 0.850 -> 0.960 (MNIST HMC FGSM) | No clear effect |
| increase number of posterior samples averaged in the expected gradient (ablation) | HMC and VI BNN · MNIST and Fashion MNIST | magnitude of expected loss gradient | Helped |
| train BNN with HMC instead of SGD deterministic training | MNIST/Fashion MNIST networks · accuracy-robustness trade-off under FGSM | robustness vs accuracy | Helped |
| use VI instead of HMC for BNN posterior | BNN · MNIST and Fashion MNIST | robustness vs accuracy | Hurt |
| increase network width (neurons) (ablation) | HMC BNN · MNIST and Fashion MNIST | robustness to FGSM | Helped |
| increase network width (ablation) | VI BNN · MNIST | robustness to FGSM | Mixed |
Closest papers by meaning
- Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network (ICLR 2019 · similarity 0.91)
- Bayesian Learning with Information Gain Provably Bounds Risk for a Robust Adversarial Defense (ICML 2022 · similarity 0.90)
- Defense Through Diverse Directions (ICML 2020 · similarity 0.89)
- BNN-DP: Robustness Certification of Bayesian Neural Networks via Dynamic Programming (ICML 2023 · similarity 0.89)
- Accurate, reliable and fast robustness evaluation (NeurIPS 2019 · similarity 0.88)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · 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.