Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting Adversarial Attacks With Momentum
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 |
|---|---|---|---|
| Add accumulated-gradient momentum to iterative FGSM | I-FGSM attacks crafted on Inc-v3 · ImageNet validation, transfer to Inc-v4 | success rate (%): 22.8 -> 48.8 | Helped |
| Add accumulated-gradient momentum to iterative FGSM | I-FGSM attacks crafted on Inc-v3 · ImageNet validation, transfer to IncRes-v2 | success rate (%): 19.9 -> 48.0 | Helped |
| Add accumulated-gradient momentum to iterative FGSM | I-FGSM attacks crafted on Inc-v3 · ImageNet validation, transfer to Res-152 | success rate (%): 16.2 -> 35.6 | Helped |
| Add accumulated-gradient momentum to iterative FGSM | I-FGSM attacks crafted on Inc-v3 · ImageNet validation, transfer to Inc-v3 ens3 | success rate (%): 7.5 -> 15.1 | Helped |
| Fuse ensemble model outputs at the logits rather than prediction or loss level (ablation) | MI-FGSM ensemble attacks · ImageNet validation, hold out Inc-v3 | success rate (%): 83.3 -> 87.9; 81.9 -> 87.9 | Helped |
| Fuse ensemble model outputs at the logits rather than prediction or loss level (ablation) | MI-FGSM ensemble attacks · ImageNet validation, hold out Inc-v4 | success rate (%): 77.4 -> 81.2; 72.5 -> 81.2 | Helped |
| Add momentum to iterative ensemble attacks | I-FGSM ensemble of six models · ImageNet validation, hold out Inc-v3 ens4 | success rate (%): 18.7 -> 40.3 | Helped |
| Use momentum iterative attacks instead of iterative attacks | Targeted ensemble attacks · ImageNet validation, targeted L∞, hold out Inc-v3 | targeted success rate (%): 9.0 -> 17.6 | Helped |
| Increase the decay factor from zero toward 1.0 (ablation) | MI-FGSM attacks crafted on Inc-v3 · ImageNet validation, black-box models | success rate (%) | Mixed |
| Increase the number of iterations (ablation) | I-FGSM attacks crafted on Inc-v3 · ImageNet validation, black-box models | success rate (%) | Hurt |
| Increase the number of iterations (ablation) | MI-FGSM attacks crafted on Inc-v3 · ImageNet validation, black-box models | success rate (%) | No clear effect |
| Increase perturbation size (ablation) | MI-FGSM attacks crafted on Inc-v3 · ImageNet validation, transfer to Res-152 | success rate (%) | Helped |
Closest papers by meaning
- On the Convergence of an Adaptive Momentum Method for Adversarial Attacks (AAAI 2024 · similarity 0.91)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.87)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.87)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.86)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.85)
- Mind the Gap: Detecting Black-box Adversarial Attacks in the Making through Query Update Analysis (CVPR 2025 · 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.