Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Building More Robust Models with Frequency Bias
Qingwen Bu, Dong Huang, Heming Cui
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| insert FPCM into AWP adversarial training | AWP · CIFAR-10 WRN-34-10 | AutoAttack robust accuracy: 53.90 -> 55.18 | Improved |
| insert FPCM into PGD-AT | PGD-AT · CIFAR-10 ResNet18 | AutoAttack robust accuracy: 47.42 -> 49.26 | Improved |
| insert FPCM into TRADES | TRADES · CIFAR-10 WRN-34-10 | AutoAttack robust accuracy: 52.30 -> 52.88 | Improved |
| insert FPCM into PGD-AT on Imagenette | PGD-AT · Imagenette ResNet18 | AutoAttack robust accuracy: 59.3 -> 60.5 | Improved |
| insert FPCM into PGD-AT on CIFAR-100 | PGD-AT · CIFAR-100 ResNet18 | AutoAttack robust accuracy: 22.37 -> 25.19 | Improved |
| insert FPCM into TRADES on CIFAR-100 WRN-34-10 | TRADES · CIFAR-100 WRN-34-10 | clean accuracy: 58.43 -> 57.75 | Hurt |
| feature-level FPCM instead of input low-pass filtering (FR) | frequency-based defense · CIFAR-10 WRN-34-10 | clean accuracy and AutoAttack: clean 80.59 -> 85.15 vs FR; AA 54.35 -> 55.18 vs FR-SWA | Improved |
| weight low-frequency features more than 0.5 (alpha above 0.5) (ablation) | FPCM · CIFAR-10 ResNet18 | W-Robust: +0.37% to +1.79% | Improved |
| set alpha to 0 (only high-frequency) (ablation) | FPCM · CIFAR-10 ResNet18 | convergence | Hurt |
| learnable alpha instead of fixed (ablation) | FPCM · CIFAR-10 ResNet18 | W-Robust: +0.39% W-Robust | Improved |
| use MLP instead of 1-D conv for learnable alpha (ablation) | FPCM · CIFAR-10 ResNet18 | W-Robust: 68.31 -> 68.47 | No clear effect |
| linearly decrease cutoff frequency during training (ablation) | FPCM · CIFAR-10 ResNet18 | W-Robust: 65.23-65.82 (fixed) -> 66.04 (linear) | Improved |
Closest papers by meaning
- First Line of Defense: A Robust First Layer Mitigates Adversarial Attacks (AAAI 2025 · similarity 0.86)
- DAT: Improving Adversarial Robustness via Generative Amplitude Mix-up in Frequency Domain (NeurIPS 2024 · similarity 0.86)
- Robust Image Denoising through Adversarial Frequency Mixup (CVPR 2024 · similarity 0.85)
- Tuning Frequency Bias of State Space Models (ICLR 2025 · similarity 0.85)
- FACL-Attack: Frequency-Aware Contrastive Learning for Transferable Adversarial Attacks (AAAI 2024 · similarity 0.85)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.85)
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.
