Research map / Adversarial robustness and security / Certified and empirical robustness
Adv-BNN: Improved Adversarial Defense through Robust Bayesian Neural Network
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 |
|---|---|---|---|
| Combine adversarial training with a Bayesian neural network using stochastic weights | VGG · CIFAR-10, white-box PGD ε=0.035 | accuracy: 31.1 -> 45.4 | Helped |
| Combine adversarial training with a Bayesian neural network using stochastic weights | Model A · STL-10, white-box PGD ε=0.035 | accuracy: 27.4 -> 37.6 | Helped |
| Combine adversarial training with a Bayesian neural network using stochastic weights | VGG · ImageNet-143, white-box PGD ε=0.01 | accuracy: 23.0 -> 39.3 | Helped |
| Use Adv-BNN instead of adversarial training | VGG · CIFAR-10, clean test set (PGD ε=0) | accuracy: 80.3 -> 79.7 | Hurt |
| Use Adv-BNN instead of adversarial training | Model A · STL-10, clean test set (PGD ε=0) | accuracy: 63.2 -> 59.9 | Hurt |
| Use Adv-BNN instead of adversarial training | VGG · ImageNet-143, clean test set (PGD ε=0) | accuracy: 48.7 -> 47.3 | Hurt |
| Train a Bayesian neural network without adversarial training (ablation) | BNN · CIFAR-10 adversarial robustness | accuracy | No clear effect |
| Increase PGD attack iterations beyond 20, up to 1000 (ablation) | Adv-BNN · ImageNet-143 | accuracy | No clear effect |
| Average predictions from 10–20 sampled weight models (ablation) | Adv-BNN · ImageNet-143 prediction under perturbations | accuracy: 10${\sim}$20 | Helped |
Closest papers by meaning
- Bayesian Learning with Information Gain Provably Bounds Risk for a Robust Adversarial Defense (ICML 2022 · similarity 0.93)
- Defense Through Diverse Directions (ICML 2020 · similarity 0.92)
- Robustness of Bayesian Neural Networks to Gradient-Based Attacks (NeurIPS 2020 · similarity 0.91)
- Improving Bayesian Neural Networks by Adversarial Sampling (AAAI 2022 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · 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.