Research map / Adversarial robustness and security / Black-box and transfer attacks
Diversified Adversarial Attacks based on Conjugate Gradient Method
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 |
|---|---|---|---|
| Replace APGD's gradient update direction with conjugate-gradient directions | ACG versus APGD · RobustBench models, CIFAR-10, CIFAR-100, and ImageNet | attack success rate: 49.67 -> 55.77 | Helped |
| Use one ACG restart instead of five APGD restarts | ACG versus APGD · 64 RobustBench models | attack success rate | Helped |
| Use conjugate-gradient updates rather than gradient updates during the early search, then switch to gradient updates (ablation) | ACG search · CIFAR-10, WideResNet-28-4 | attack success rate: 55.77 -> 55.84 | Helped |
| Use gradient updates early and switch to conjugate-gradient updates at step size ε (ablation) | ACG search · CIFAR-10, WideResNet-28-4 | attack success rate: 55.77 -> 50.85 | Hurt |
| Constrain the conjugate-gradient coefficient βHS to be nonnegative (ablation) | ACG · CIFAR-10 (ε=8/255), WideResNet-28-4 | attack success rate: 49.67 -> 50.68; 55.77 -> 50.68 | Mixed |
| Use DLR loss instead of CW loss (ablation) | ACG and APGD · CIFAR-10 and ImageNet | attack success rate: +6.30 | Helped |
| Extend ACG from five 100-iteration restarts to one 500-iteration deterministic run (ablation) | ACG · CIFAR-10, CIFAR-100, and ImageNet | attack success rate: 55.77 -> 55.32 | No clear effect |
Closest papers by meaning
- On the Diversity of Adversarial Ensemble Learning (ICML 2025 · similarity 0.85)
- Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path (AAAI 2025 · similarity 0.84)
- DiffAdvMAP: Flexible Diffusion-Based Framework for Generating Natural Unrestricted Adversarial Examples (ICML 2025 · similarity 0.84)
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 2025 · similarity 0.84)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.84)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.84)
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.