Research map / Adversarial robustness and security / Black-box and transfer attacks
Making Adversarial Examples More Transferable and Indistinguishable
Junhua Zou, Yexin Duan, Boyu Li, Wu Zhang, Yu Pan, Zhisong Pan
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 the sign function with tanh (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean perturbation: 10.46 -> 9.14 | Improved |
| Replace the sign function with tanh (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean success rate (%): 82.0 -> 82.4 | Improved |
| Use Adam instead of the baseline gradient update (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean success rate (%): 82.0 -> 83.6 | Improved |
| Use Adam instead of the baseline gradient update (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean perturbation: 10.46 -> 9.20 | Improved |
| Use tanh together with Adam (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean perturbation: 10.46 -> 7.86 | Improved |
| Use tanh together with Adam (ablation) | TI-DIM · Inc-v3 attack; six classic defenses | mean success rate (%): 82.0 -> 83.1 | Improved |
| Use smaller Gaussian kernels (ablation) | TI-DIM + tanh + Adam · Inc-v3 attack; six classic defenses | mean success rate (%): 83.1 -> 86.5 | Improved |
| Use increasing step sizes (ablation) | TI-DIM + tanh + Adam + smaller kernels · Inc-v3 attack; six classic defenses | mean success rate (%): 86.5 -> 88.0 | Improved |
| Use increasing step sizes (ablation) | TI-DIM + tanh + Adam + smaller kernels · Inc-v3 attack; six classic defenses | mean perturbation: 7.82 -> 8.11 | Hurt |
| Replace TI-DIM with TI-DI-AITM | Ensemble attacks · Six classic defenses | mean attack success rate (%): 82.0 -> 88.0 | Improved |
| Replace NI-TI-DIM with NI-TI-DI-AITM | Ensemble attacks · Six classic defenses | mean attack success rate (%): 83.7 -> 89.3 | Improved |
| Replace NI-TI-DIM with NI-TI-DI-AITM | Ensemble attacks · Three advanced defenses | mean attack success rate (%): 75.1 -> 82.7 | Improved |
Closest papers by meaning
- Nesterov Accelerated Gradient and Scale Invariance for Adversarial Attacks (ICLR 2020 · similarity 0.91)
- Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attacks (CVPR 2019 · similarity 0.90)
- On the Convergence of an Adaptive Momentum Method for Adversarial Attacks (AAAI 2024 · similarity 0.87)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.87)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.87)
- DiffAdvMAP: Flexible Diffusion-Based Framework for Generating Natural Unrestricted Adversarial Examples (ICML 2025 · 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.
