Research map / Adversarial robustness and security / Certified and empirical robustness
On the Structural Sensitivity of Deep Convolutional Networks to the Directions of Fourier Basis Functions
Yusuke Tsuzuku, Issei Sato
Check 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 selected Fourier basis perturbation instead of random noise | CNN classifiers · CIFAR10 | fool ratio: 5.0 -> 63.3; 8.1 -> 82.3; 6.8 -> 72.2; 5.4 -> 50.7 | Improved |
| Use a selected Fourier basis perturbation instead of random noise | CNN classifiers · CIFAR100 | fool ratio: 13.0 -> 83.4; 26.4 -> 93.7; 25.9 -> 95.8; 22.5 -> 72.3 | Improved |
| Use a selected Fourier basis perturbation instead of random noise | CNN classifiers · MNIST | fool ratio: 0.1 -> 0.4; 55.8 -> 90.2; 0.0 -> 0.1; 0.1 -> 0.2 | Mixed |
| Use a selected Fourier basis perturbation instead of random noise | CNN classifiers · Fashion MNIST | fool ratio: 5.4 -> 12.5; 11.4 -> 48.1; 8.3 -> 83.7; 12.6 -> 56.9 | Improved |
| Use a selected Fourier basis perturbation instead of random noise | CNN classifiers · SVHN | fool ratio: 3.1 -> 64.9; 5.2 -> 90.8; 0.0 -> 0.0; 4.6 -> 50.5 | Mixed |
| Replace SFA's Fourier basis with its signed version (ablation) | SFA on ILSVRC2015 · ILSVRC2015 | fool ratio: 34.6 -> 44.1; 38.7 -> 40.1; 49.7 -> 53.3; 36.8 -> 39.5 | Mixed |
| Increase perturbation size from 10/255 to 20/255 | SFA on ILSVRC2015 · ILSVRC2015 | fool ratio: 34.6 -> 62.3; 38.7 -> 68.5; 49.7 -> 76.3; 36.8 -> 63.5 | Improved |
| Use a Fourier basis perturbation rather than a standard pixel basis direction | CNNs · Various datasets | fool ratio | Improved |
| Use convolutional networks rather than an MLP | Fourier-direction sensitivity · Various datasets | fool ratio | Improved |
| Use a single Fourier basis perturbation on MNIST | CNN classifiers · MNIST | fool ratio | Hurt |
Closest papers by meaning
- Universal Adversarial Perturbations (CVPR 2017 · similarity 0.89)
- Lower bounds on the robustness to adversarial perturbations (NeurIPS 2017 · similarity 0.89)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.88)
- Sensitivity and Generalization in Neural Networks: an Empirical Study (ICLR 2018 · similarity 0.88)
- Defense Against Universal Adversarial Perturbations (CVPR 2018 · similarity 0.88)
- Verification of Neural Networks Against Convolutional Perturbations via Parameterised Kernels (AAAI 2025 · similarity 0.88)
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.
