Research map / Adversarial robustness and security / Certified and empirical robustness
How Sampling Impacts the Robustness of Stochastic Neural Networks
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 |
|---|---|---|---|
| Increase the number of samples used to calculate the attack | Stochastic neural networks under gradient-based attacks · FashionMNIST and CIFAR10 | adversarial accuracy | Hurt |
| Use more samples during inference (ablation) | BNN · FashionMNIST, FGM attack | adversarial accuracy: 35.76 \pm 0.92 -> 36.50 \pm 0.39 | No clear effect |
| Use more samples during inference (ablation) | Infinite mixture (IM) · FashionMNIST, FGM attack | adversarial accuracy: 44.67 \pm 1.19 -> 47.96 \pm 0.24 | No clear effect |
| Use more samples during inference (ablation) | SIN 0.05 · FashionMNIST, FGM attack | adversarial accuracy: 36.15 \pm 0.79 -> 35.91 \pm 0.23 | No clear effect |
| Use more samples during inference (ablation) | SIN 0.1 · FashionMNIST, FGM attack | adversarial accuracy: 40.63 \pm 0.79 -> 39.67 \pm 0.29 | No clear effect |
| Use more samples during inference (ablation) | ResNet dropout 0.3 · CIFAR10, FGM attack | adversarial accuracy: 42.64 \pm 0.39 -> 42.82 \pm 0.17 | No clear effect |
| Use more samples during inference (ablation) | ResNet dropout 0.6 · CIFAR10, FGM attack | adversarial accuracy: 47.18 \pm 0.70 -> 47.02 \pm 0.20 | No clear effect |
| Increase stochasticity, represented by higher prediction variance | SNNs with otherwise similar architecture and training · FashionMNIST and CIFAR10 | adversarial accuracy | Helped |
| Use a CW loss on averaged logits instead of the original attack objective | Infinite mixture (IM) · FashionMNIST, FGM attack | adversarial accuracy | Hurt |
| Use a CW loss on averaged logits instead of the original attack objective | BNN · FashionMNIST, FGM attack | adversarial accuracy | No clear effect |
| Increase attack sample count | SNNs under 100-iteration PGD · FashionMNIST and CIFAR10 | adversarial accuracy | Hurt |
| Increase attack sample count | SNNs under FGM with an ℓ∞ constraint · FashionMNIST and CIFAR10 | adversarial accuracy | Hurt |
Closest papers by meaning
- Theoretical evidence for adversarial robustness through randomization (NeurIPS 2019 · similarity 0.91)
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.89)
- Enhancing Adversarial Robustness in SNNs with Sparse Gradients (ICML 2024 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.87)
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 2024 · 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.