Research map / Adversarial robustness and security / Attacks on text and graphs
Enhancing Adversarial Robustness via Score-Based Optimization
Boya Zhang, Weijian Luo, Zhihua Zhang
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 |
|---|---|---|---|
| use ScoreOpt-N instead of ScoreOpt-O (ablation) | ScoreOpt · CIFAR-10 BPDA+EOT | robust accuracy (%): 82.85 -> 90.59 | Improved |
| increase diffusion purification denoising steps from 1 to 80 (ablation) | Diffusion purification · CIFAR-100 BPDA+EOT | robust accuracy (%): 40.03 -> 47.85 | No clear effect |
| replace multi-step diffusion purification with ScoreOpt-O | ScoreOpt · CIFAR-100 BPDA+EOT | robust accuracy (%): 48.83 -> 66.99 | Improved |
| use approximate one-shot-denoising gradient instead of exact gradient for PGD+EOT (ablation) | ScoreOpt · CIFAR-10 512 subset l_2 | attack success rate (%): 33.99 -> 34.11 | No clear effect |
| combine ScoreOpt with adversarially trained classifier | ScoreOpt · CIFAR-10 PGD+EOT l_inf | robust accuracy (%): 71.48 -> 72.57 | Improved |
| combine ScoreOpt with adversarially trained classifier | ScoreOpt · CIFAR-10 PGD+EOT l_2 | robust accuracy (%): 87.11 -> 89.06 | Improved |
| increase number of optimization steps (ablation) | ScoreOpt · CIFAR-10 BPDA+EOT | standard and robust accuracy | Improved |
| vary number of EOT replicates (ablation) | ScoreOpt · CIFAR-10 PGD+EOT and BPDA+EOT | accuracy | No clear effect |
| use WRN-70-16 classifier instead of WRN-28-10 (ablation) | ScoreOpt-N · CIFAR-10 BPDA+EOT | robust accuracy (%): 90.59 -> 91.60 | Improved |
| change noise level to [0.60,0.60] from [0.50,0.50] (ablation) | ScoreOpt-N · CIFAR-10 BPDA+EOT | robust accuracy (%): 90.59 -> 89.26 | Hurt |
| change noise level to [0.40,0.40] from [0.50,0.50] (ablation) | ScoreOpt-N · CIFAR-10 BPDA+EOT | robust accuracy (%): 90.59 -> 92.19 | Improved |
| change noise level to [0.30,0.50] from [0.40,0.60] (ablation) | ScoreOpt-O · CIFAR-10 BPDA+EOT | robust accuracy (%): 82.85 -> 79.30 | Hurt |
Closest papers by meaning
- Towards Understanding the Robustness of Diffusion-Based Purification: A Stochastic Perspective (ICLR 2025 · similarity 0.87)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.86)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 2024 · similarity 0.86)
- Adversarial Score Distillation: When score distillation meets GAN (CVPR 2024 · similarity 0.86)
- GREAT Score: Global Robustness Evaluation of Adversarial Perturbation using Generative Models (NeurIPS 2024 · similarity 0.86)
- Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization (ICLR 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.
