Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Bayesian Neural Networks by Spectral Expectation Bound Regularization
Check 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 |
|---|---|---|---|
| Add SEBR (ablation) | Bayesian MLP · MNIST, FGSM medium (ℓ∞ norm 0.16) | accuracy: 8.97 ± 0.28 -> 43.69 ± 5.92 | Helped |
| Add SEBR (ablation) | Bayesian CNN · MNIST, PGD medium (ℓ∞ norm 0.08) | accuracy: 36.53 ± 5.85 -> 49.20 ± 10.75 | Helped |
| Add SEBR (ablation) | Bayesian MLP · Fashion-MNIST, PGD medium (ℓ∞ norm 0.1) | accuracy: 19.18 ± 1.01 -> 29.67 ± 1.22 | Helped |
| Add SEBR (ablation) | Bayesian MLP · MNIST, clean | accuracy: 97.05 ± 0.38 -> 96.83 ± 0.48 | Hurt |
| Add SEBR (ablation) | Adversarially trained Bayesian MLP · MNIST, FGSM large (ℓ∞ norm 0.3) | accuracy: 9.94 ± 0.13 -> 33.09 ± 8.23 | Helped |
| Use the analytic upper bound instead of directly regularizing expected spectral norms | Bayesian neural network training · Training epoch | average time per epoch: 1654.8 (s) -> 410.5 (s) | Helped |
| Increase the SEBR regularization factor beyond a suitable value (ablation) | Bayesian neural network · Adversarial examples | accuracy | Hurt |
Closest papers by meaning
- Robust Learning of Fixed-Structure Bayesian Networks (NeurIPS 2018 · similarity 0.88)
- Robust Learning of Fixed-Structure Bayesian Networks in Nearly-Linear Time (ICLR 2021 · similarity 0.87)
- Defense Through Diverse Directions (ICML 2020 · similarity 0.87)
- Bayesian Learning with Information Gain Provably Bounds Risk for a Robust Adversarial Defense (ICML 2022 · similarity 0.87)
- Improving Bayesian Neural Networks by Adversarial Sampling (AAAI 2022 · similarity 0.87)
- PAC-Bayesian Spectrally-Normalized Bounds for Adversarially Robust Generalization (NeurIPS 2023 · similarity 0.87)
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.