Research map / Adversarial robustness and security / Attacks on text and graphs
Spatial-frequency channels, shape bias, and adversarial robustness
Ajay Subramanian, Elena Sizikova, Najib J. Majaj, Denis G. Pelli
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 |
|---|---|---|---|
| Use a narrower channel bandwidth | Neural networks evaluated on 16-way ImageNet categorization · 16-way ImageNet cue-conflict categorization | shape bias: 29% | Improved |
| Increase channel bandwidth and peak noise sensitivity | Adversarially trained networks · ImageNet validation with PGD whitebox attack | whitebox accuracy: 66% variance explained by the three channel properties together | Improved |
| Increase the amount of adversarial training | ResNet-50 · ImageNet validation with PGD whitebox attack | whitebox accuracy | Improved |
| Increase the amount of adversarial training | ResNet-50 · ImageNet validation set | accuracy | Hurt |
| Increase the amount of adversarial training | ResNet-50 · 16-way ImageNet categorization with narrowband noise | channel bandwidth | Hurt |
| Use more than a one-octave-wide channel | Neural networks · 16-way ImageNet categorization with narrowband noise | channel bandwidth: 2.30 to 5.98, compared with 1.00 octave | Hurt |
| Use a weaker PGD evaluation attack, 20-step PGD with ||L∞||=4/255 | Adversarially trained networks · ImageNet validation whitebox robustness | correlation between channel bandwidth and whitebox accuracy | No clear effect |
Closest papers by meaning
- High-Frequency Component Helps Explain the Generalization of Convolutional Neural Networks (CVPR 2020 · similarity 0.83)
- Amplitude-Phase Recombination: Rethinking Robustness of Convolutional Neural Networks in Frequency Domain (ICCV 2021 · similarity 0.83)
- Learning in the Frequency Domain (CVPR 2020 · similarity 0.83)
- Finding Biological Plausibility for Adversarially Robust Features via Metameric Tasks (ICLR 2022 · similarity 0.83)
- Spectral Bias in Practice: The Role of Function Frequency in Generalization (NeurIPS 2022 · similarity 0.82)
- Contour Integration Underlies Human-Like Vision (ICML 2025 · similarity 0.81)
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.
