Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting Adversarial Transferability via Negative Hessian Trace Regularization
Yunfei Long, Zilin Tian, Liguo Zhang, Huosheng Xu
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 |
|---|---|---|---|
| replace first-order flatness regularizer PGN with negative Hessian trace regularization | MI-FGSM, Inc-v3 surrogate · ImageNet-compatible, avg over 5 targets | attack success rate (%): 64.1 -> 69.7 | Improved |
| penalize negative Hessian trace rather than minimizing the trace directly | NHTR objective · adversarial example flatness | Hessian trace: 9.3 (PGN) -> 5.2 | Improved |
| add NHTR to MI-FGSM with DIM input transformation (ablation) | MI-FGSM + DIM · Inc-v3 to Res-152 | attack success rate (%): 42.3 -> 78.7 | Improved |
| add NHTR to MI-FGSM with Admix (ablation) | MI-FGSM + Admix · Inc-v3 to Res-152 | attack success rate (%): 44.8 -> 79.6 | Improved |
| add NHTR to MI-FGSM with BSR (ablation) | MI-FGSM + BSR · Inc-v3 to Res-152 | attack success rate (%): 65.6 -> 84.7 | Improved |
| use NHTR in model-ensemble setting instead of PGN | ensemble surrogate · adversarially trained targets avg | attack success rate (%): 80.0 -> 86.3 | Improved |
| use NHTR instead of PGN against defense methods RS | Inc-v3 surrogate · RS defense | attack success rate (%): 7.1 -> 11.3 | Improved |
| use NHTR instead of PGN for computation cost | attack generation · 10 images 299x299, single 4090 | time (s): 21.1 -> 33.3 | Hurt |
| set balance coefficient delta to 0.3 (ablation) | NHTR · black-box transferability | attack success rate | Improved |
| set neighborhood radius xi to 3 times epsilon (ablation) | NHTR · black-box transferability | attack success rate | Improved |
Closest papers by meaning
- Boosting Adversarial Transferability by Achieving Flat Local Maxima (NeurIPS 2023 · similarity 0.92)
- Enhancing Adversarial Transferability With Cost-Efficient Landscape Flattening (TPAMI 2026 · similarity 0.90)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · similarity 0.87)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.87)
- Transferability Bound Theory: Exploring Relationship between Adversarial Transferability and Flatness (NeurIPS 2024 · similarity 0.87)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.87)
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.
