Research map / Adversarial robustness and security / Certified and empirical robustness
Stochastic Activation Pruning for Robust Adversarial Defense
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 SAP-100 to the dense classifier | Dense ResNet-20 · CIFAR-10, FGSM λ=1, 2, 4 | accuracy: 12.2%, 16.3% and 12.8% absolute increase | Helped |
| Apply SAP-100 to the dense classifier | Dense ResNet-20 · CIFAR-10, clean | accuracy: 6.5% absolute decrease | Hurt |
| Apply SAP-100 to the dense classifier | Dense ResNet-20 · CIFAR-10, FGSM λ=16 | accuracy: 5.2% absolute decrease | Hurt |
| Add SAP-100 to adversarial training | Adversarially trained ResNet-20 · CIFAR-10, adversarial examples λ=16 and 32 | accuracy: 7.8% and 7.9% absolute increase | Helped |
| Use stochastic activation pruning rather than ordinary dropout | Pretrained ResNet-20 · CIFAR-10, adversarial examples | accuracy | Helped |
| Apply dropout to the pretrained model (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | No clear effect |
| Train the model using dropout (ablation) | ResNet-20 · CIFAR-10, adversarial examples | accuracy | No clear effect |
| Add zero-mean Gaussian noise to weights (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | Hurt |
| Scale weights with multiplicative Gaussian noise (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | No clear effect |
| Add Gaussian noise to activations (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | Hurt |
| Scale activations with multiplicative Gaussian noise (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | No clear effect |
| Prune weights deterministically by magnitude (ablation) | Dense ResNet-20 · CIFAR-10, adversarial examples | accuracy | No clear effect |
Closest papers by meaning
- On the Limitations of Stochastic Pre-processing Defenses (NeurIPS 2022 · similarity 0.90)
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.88)
- Adversary Aware Optimization for Robust Defense (NeurIPS 2025 · similarity 0.88)
- Are Adversarial Examples Created Equal? A Learnable Weighted Minimax Risk for Robustness under Non-uniform Attacks (AAAI 2021 · similarity 0.88)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.87)
- Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection (NeurIPS 2026 · 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.