Research map / Adversarial robustness and security / Attacks on text and graphs
Sparse Adversarial Perturbations for Videos
Xingxing Wei, Jun Zhu, Sha Yuan, Hang Su
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 |
|---|---|---|---|
| Increase the proportion of clean frames to 97.5%, leaving one perturbed frame (ablation) | Sparse adversarial perturbations · UCF101 test | fooling rate: 100% -> 59.7% | Hurt |
| Increase sparsity from 80% to 90% (ablation) | Sparse adversarial perturbations · UCF101 test | fooling rate: 100% -> 91.8% | Hurt |
| Use l2,1 regularization instead of l2 regularization (ablation) | Video perturbation optimization · UCF101 | temporal sparsity | Improved |
| Use l2 regularization rather than l2,1 when perturbing selected frames for propagation to the rest (ablation) | Shortened-video propagation attack · UCF101 | fooling rate | Improved |
| Add a temporal mask to keep frames outside the selected top 8 clean (ablation) | Sparse adversarial perturbations · UCF101 videos | frame-level misclassification | Improved |
| Replace VanillaRNN with LSTM or GRU (ablation) | CNN+RNN video attack · UCF101 | fooling rate: 84.1% -> 100% (LSTM); 81.8% -> 100% (GRU) | Improved |
| Remove the RNN layer and use CNN + Average Pooling (ablation) | CNN+RNN video attack · UCF101 | fooling rate: 100% -> 87.3% | Hurt |
| Use perturbations trained on the training videos on unseen test videos | Universal perturbations · UCF101 test | fooling rate: 100% (training set); 95.2% (testing set) | Improved |
| Compute perturbations on fewer frames of a shortened video (ablation) | Propagation-based attack · UCF101, one optimization iteration | computation time: 2.853s -> 0.0947s | Improved |
Closest papers by meaning
- Sparse Adversarial Attack via Perturbation Factorization (ECCV 2020 · similarity 0.90)
- From Pretrain to Pain: Adversarial Vulnerability of Video Foundation Models Without Task Knowledge (AAAI 2026 · similarity 0.88)
- Over-the-Air Adversarial Flickering Attacks Against Video Recognition Networks (CVPR 2021 · similarity 0.88)
- Transferable Structural Sparse Adversarial Attack Via Exact Group Sparsity Training (CVPR 2024 · similarity 0.87)
- Contrasting Adversarial Perturbations: The Space of Harmless Perturbations (AAAI 2025 · similarity 0.87)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.86)
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.
