Research map / Adversarial robustness and security / Attacks on text and graphs
LAS-AT: Adversarial Training With Learnable Attack Strategy
Check 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 learnable attack strategies to AWP | AWP with WRN34-10 · CIFAR-10 | clean accuracy: 85.57 -> 87.74 | Helped |
| Add learnable attack strategies to AWP | AWP with WRN34-10 · CIFAR-10 PGD-10 | robust accuracy: 58.92 -> 61.09 | Helped |
| Add learnable attack strategies to AWP | AWP with WRN34-10 · CIFAR-10 AA | robust accuracy: 53.90 -> 55.52 | Helped |
| Add learnable attack strategies to AWP | AWP with PreActResNet18 · Tiny ImageNet | clean accuracy: 41.48 -> 45.26 | Helped |
| Add learnable attack strategies to AWP | AWP with WRN34-20 · CIFAR-100 | clean accuracy: 62.55 -> 67.31 | Helped |
| Add robustness-evaluation loss L2 (ablation) | LAS-PGD-AT · CIFAR-10 PGD-10 | robust accuracy: 53.88 -> 53.98 | Helped |
| Add robustness-evaluation loss L2 (ablation) | LAS-PGD-AT · CIFAR-10 | clean accuracy: 81.83 -> 81.54 | Hurt |
| Add clean-sample loss L3 (ablation) | LAS-PGD-AT · CIFAR-10 | clean accuracy: 81.83 -> 81.90 | Helped |
| Add clean-sample loss L3 (ablation) | LAS-PGD-AT · CIFAR-10 AA | robust accuracy: 49.06 -> 49.20 | Hurt |
| Use both L2 and L3 (ablation) | LAS-PGD-AT · CIFAR-10 | clean accuracy: 81.83 -> 82.3 | Helped |
| Use both L2 and L3 (ablation) | LAS-PGD-AT · CIFAR-10 PGD-10 | robust accuracy: 53.88 -> 54.29 | Helped |
| Use both L2 and L3 (ablation) | LAS-PGD-AT · CIFAR-10 AA | robust accuracy: 49.06 -> 49.89 | Helped |
Closest papers by meaning
- Adversarial Attack Generation Empowered by Min-Max Optimization (NeurIPS 2021 · similarity 0.89)
- Adversarial Distributional Training for Robust Deep Learning (NeurIPS 2020 · similarity 0.88)
- Understanding and Increasing Efficiency of Frank-Wolfe Adversarial Training (CVPR 2022 · similarity 0.88)
- Revisiting and Advancing Fast Adversarial Training Through The Lens of Bi-Level Optimization (ICML 2022 · similarity 0.88)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.87)
- Clarifying the Behavior and the Difficulty of Adversarial Training (AAAI 2024 · 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.