Research map / Adversarial robustness and security / Certified and empirical robustness
With Friends Like These, Who Needs Adversaries?
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 |
|---|---|---|---|
| Downsample input images before upsampling to original size | AlexNet · ImageNet test | accuracy: 57.75 at 227 -> 48.10 at 100 | Hurt |
| Apply original-network DeepFool perturbations after input downsampling | AlexNet with downsampling · ImageNet test | DeepFool fooling rate: 100.00 at 227; 32.75, 35.55, 36.15, 41.65 at 200, 150, 120, 100 | Mixed |
| Increase original-network DeepFool perturbation norm by scaling f | AlexNet with downsampling · ImageNet test | DeepFool fooling rate: 32.75 -> 97.65 at 200; 35.55 -> 95.65 at 150; 36.15 -> 92.90 at 120; 41.65 -> 88.30 at 100 | Helped |
| Restrict DeepFool optimization to high-curvature classification directions | DCN classifiers · Image classification | adversarial perturbation norm | Helped |
| Project inputs onto the most curved directions rather than flat directions | DCN classifiers · Image classification | classification accuracy | Helped |
| Use positively or negatively curved directions instead of flat directions | DCN classifiers · Image classification | classification accuracy | Helped |
| Downsample images to reduce effective input dimensionality | AlexNet · ImageNet test | robustness to transfer attack: 100.00 -> 32.75 at d_low=200; 100.00 -> 41.65 at d_low=100 (f=1) | Mixed |
Closest papers by meaning
- First-Order Adversarial Vulnerability of Neural Networks and Input Dimension (ICML 2019 · similarity 0.92)
- Robust Feature-Level Adversaries are Interpretability Tools (NeurIPS 2022 · similarity 0.90)
- Benchmarking Adversarial Robustness on Image Classification (CVPR 2020 · similarity 0.89)
- Universal Adversarial Perturbations (CVPR 2017 · similarity 0.89)
- Doppelgangers and Adversarial Vulnerability (CVPR 2025 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 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.