Research map / Adversarial robustness and security / Certified and empirical robustness
Hold me tight! Influence of discriminative features on deep network boundaries
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 |
|---|---|---|---|
| train on low-pass filtered CIFAR-10 instead of standard (ablation) | DenseNet-121 · CIFAR-10 test | accuracy (%): 93.03% -> 89.67% | Hurt |
| train on low-pass CIFAR-10 and test on low-pass data versus CIFAR-10-trained net (ablation) | DenseNet-121 · CIFAR-10 low-pass test | accuracy (%): 27.45% -> 88.51% | Helped |
| train on low-pass MNIST (14x14 bandwidth) (ablation) | LeNet · MNIST test | accuracy | No clear effect |
| flip DCT spectrum of training data (ablation) | ResNet-50 · ImageNet flipped | accuracy (%): 0.292% -> 68.12% | Helped |
| increase small feature size eps from below 0.2 to above 0.5 in synthetic T2 (ablation) | MLP · synthetic T2 (rho=20) | boundary type | Mixed |
| add weight decay 1e-3 (ablation) | MLP · synthetic T2 eps<0.5 | final train accuracy: 50% | Hurt |
| adversarial training versus standard training | DenseNet-121 · CIFAR-10 margin in high frequencies | margin | Helped |
| fine-tune on 100 adversarially perturbed training samples (ablation) | ResNet-18 · CIFAR-10 | test accuracy (%): 90% -> 90% | No clear effect |
Closest papers by meaning
- Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness (ICLR 2023 · similarity 0.91)
- Decision Boundary Analysis of Adversarial Examples (ICLR 2018 · similarity 0.90)
- Empirical Study of the Topology and Geometry of Deep Networks (CVPR 2018 · similarity 0.89)
- Robust Models Are More Interpretable Because Attributions Look Normal (ICML 2022 · similarity 0.89)
- Boundary thickness and robustness in learning models (NeurIPS 2020 · similarity 0.89)
- Reducing Excessive Margin to Achieve a Better Accuracy vs. Robustness Trade-off (ICLR 2022 · 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.