Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Robustness via Random Projection Filters
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 |
|---|---|---|---|
| replace part of convolutional filters with random projection filters | ResNet-18 · CIFAR-10 | AutoAttack accuracy: 63.78 -> 64.38 | Helped |
| replace part of convolutional filters with random projection filters | ResNet-18 · CIFAR-100 | AutoAttack accuracy: 38.13 -> 42.88 | Helped |
| replace part of convolutional filters with random projection filters | WideResNet-34-10 · CIFAR-10 | AutoAttack accuracy: 60.55 -> 68.71 | Helped |
| replace part of convolutional filters with random projection filters | ResNet-50 · ImageNet | PGD50 accuracy: 47.00 -> 55.41 | Helped |
| add random projection filters in the first layer rather than the first block (ablation) | ResNet-18 · CIFAR-10 | natural accuracy: 73.76 -> 83.79 | Helped |
| add random projection filters in the first layer rather than the first block (ablation) | ResNet-18 · CIFAR-10 | PGD accuracy: 53.80 -> 61.27 | Helped |
| move random projection filters from the first layer to deeper layers (ablation) | ResNet-18 · CIFAR-10 | AutoAttack accuracy: 63.57 -> 53.39 -> 48.06 | Hurt |
| increase the random projection filter ratio from 0.75 to 1.0 (ablation) | ResNet-18 · CIFAR-10 | natural accuracy: 83.79 -> 73.60 | Hurt |
| reduce the random projection filter ratio to 0.1 (ablation) | ResNet-18 · CIFAR-10 | PGD accuracy: 61.27 -> 58.26 | Hurt |
| increase weight decay on trainable first-layer filters from the standard setting to 1 × 10−2 (ablation) | ResNet-18 · CIFAR-10 | AutoAttack accuracy: 60.04 -> 64.38 | Helped |
| increase weight decay on trainable first-layer filters to 1 × 10−1 rather than 1 × 10−2 (ablation) | ResNet-18 · CIFAR-10 | AutoAttack accuracy: 64.38 -> 63.12 | Hurt |
Closest papers by meaning
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.90)
- Adversarial Robustness through Random Weight Sampling (NeurIPS 2023 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · similarity 0.87)
- First Line of Defense: A Robust First Layer Mitigates Adversarial Attacks (AAAI 2025 · similarity 0.86)
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 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.