Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Examples that Fool both Computer Vision and Time-Limited Humans
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 |
|---|---|---|---|
| Use targeted perturbations optimized across an ensemble of models | ImageNet images and time-limited human observers · Time-limited human classification, pets, hazard, and vegetables | probability of choosing targeted class | Helped |
| Add ensemble-targeted adversarial perturbations to class-preserving images | Time-limited human classification · Pets and vegetables; hazard also showed the reported trend | classification accuracy | Hurt |
| Use image-specific adversarial perturbations rather than vertically flipped perturbations | Time-limited human classification · Pets, vegetables, and hazard | classification accuracy | Hurt |
| Apply ensemble-targeted perturbations to images where neither response option is correct | Time-limited human observers · False condition across pets, hazard, and vegetables | probability of choosing targeted class | Helped |
| Add retinal preprocessing that approximates eccentricity-dependent visual resolution (ablation) | Adversarial attack transfer to humans · Supplementary visual comparison | perceptual similarity to target class | Helped |
| Use more models in the adversarial optimization ensemble (ablation) | Adversarial attack transfer to humans · Supplementary visual comparison | perceptual similarity to target class | Helped |
| Use a smaller perturbation size (ablation) | Adversarial attack transfer to humans · Supplementary comparison of perturbation sizes | perceptual similarity to target class | No clear effect |
Closest papers by meaning
- Adversarial Examples Are Not Bugs, They Are Features (NeurIPS 2019 · similarity 0.90)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.88)
- Doppelgangers and Adversarial Vulnerability (CVPR 2025 · similarity 0.88)
- Towards Imperceptible and Robust Adversarial Example Attacks Against Neural Networks (AAAI 2018 · similarity 0.87)
- Natural Adversarial Examples (CVPR 2021 · similarity 0.87)
- Constructing Semantics-Aware Adversarial Examples with a Probabilistic Perspective (NeurIPS 2024 · similarity 0.85)
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.