Research map / Adversarial robustness and security / Certified and empirical robustness
FrequencyLowCut Pooling – Plug & Play against Catastrophic Overfitting
Julia Grabinski, Steffen Jung, Janis Keuper, Margret Keuper
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 |
|---|---|---|---|
| replace strided/standard pooling with FLC pooling in FGSM AT | PRN-18 · CIFAR-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 0.00 -> 36.69 | Improved |
| replace pooling with FLC pooling in FGSM AT | WRN-28-10 · CIFAR-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 0.09 -> 38.37 | Improved |
| replace pooling with FLC pooling in FGSM AT | PRN-18 · CINIC-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 0.12 -> 49.61 | Improved |
| replace pooling with FLC pooling in FGSM AT | PRN-18 · CIFAR-100 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 11.13 -> 15.40 | Improved |
| use BlurPooling instead of baseline pooling in FGSM AT | PRN-18 · CIFAR-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 0.00 -> 0.06 | No clear effect |
| use Wavelet Pooling instead of baseline pooling in FGSM AT | PRN-18 · CIFAR-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 0.00 -> 5.92 | No clear effect |
| add FLC pooling in clean training | PRN-18 · CIFAR-10 (clean training) | PGD L2 0.5 accuracy: 6.68 -> 11.27 | Improved |
| add FLC pooling in clean training | PRN-18 · CIFAR-10 (clean training) | clean accuracy: 95.08 -> 94.66 | Hurt |
| add FLC pooling to FGSM AT versus FGSM with early stopping | PRN-18 · CIFAR-10 (FGSM AT) | AutoAttack Linf 8/255 accuracy: 3.76 -> 36.69 | Improved |
| add FLC pooling instead of BlurPooling for shift consistency | PRN-18 · CIFAR-10 (shift) | consistency under shift: 86.19 -> 94.46 | Improved |
| add high-frequency path to FLC pooling (ablation) | FLC pooling · CIFAR-10 (FGSM AT) | PGD Linf 8/255 accuracy: 38.41 -> 38.02 | No clear effect |
| add original-path pooling to FLC pooling (ablation) | FLC pooling · CIFAR-10 (FGSM AT) | PGD Linf 8/255 accuracy: 38.41 -> 38.30 | No clear effect |
Closest papers by meaning
- Fix Your Downsampling ASAP! Aliasing and Sinc Artifact Free Pooling in the Fourier Domain (IJCV 2026 · similarity 0.91)
- S3Pool: Pooling With Stochastic Spatial Sampling (CVPR 2017 · similarity 0.87)
- Making Convolutional Networks Shift-Invariant Again (ICML 2019 · similarity 0.87)
- Delving Deeper into Anti-Aliasing in ConvNets (IJCV 2022 · similarity 0.86)
- Group Downsampling with Equivariant Anti-aliasing (ICLR 2025 · similarity 0.86)
- Beyond Skip Connection: Pooling and Unpooling Design for Elimination Singularities (AAAI 2025 · similarity 0.84)
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.
