Research map / Adversarial robustness and security / Certified and empirical robustness
Improving the Generalization of Adversarial Training with Domain Adaptation
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 |
|---|---|---|---|
| add unsupervised domain adaptation (UDA) to standard adversarial training (ablation) | SAT / ATDA · various datasets and attacks | average accuracy (%) | Helped |
| add supervised domain adaptation (SDA) margin loss to standard adversarial training (ablation) | SAT / ATDA · various datasets and attacks | average accuracy (%) | Mixed |
| combine UDA and SDA with standard adversarial training (ablation) | ATDA · various datasets and attacks | average accuracy (%) | Helped |
| add SDA to SAT+UDA (ablation) | ATDA · various datasets and attacks | average accuracy (%) | Helped |
| use ATDA rather than SAT | Fashion-MNIST · Fashion-MNIST white-box PGD | accuracy (%): 7.4 -> 68.6 | Helped |
| use ATDA rather than SAT | SVHN · SVHN white-box PGD | accuracy (%): 44.4 -> 53.2 | Helped |
| use ATDA rather than SAT | CIFAR-10 · CIFAR-10 white-box PGD | accuracy (%): 49.5 -> 58.1 | Helped |
| use ATDA rather than SAT | CIFAR-100 · CIFAR-100 white-box PGD | accuracy (%): 18.0 -> 26.2 | Helped |
| use ATDA rather than SAT | Fashion-MNIST · Fashion-MNIST clean test | accuracy (%): 90.9 -> 85.5 | Hurt |
| use ATDA rather than SAT | SVHN · SVHN clean test | accuracy (%): 86.6 -> 82.9 | Hurt |
| use ATDA rather than SAT | CIFAR-10 · CIFAR-10 clean test | accuracy (%): 86.2 -> 84.8 | Hurt |
| add domain adaptation to PGD adversarial training | PATDA · CIFAR-10 white-box PGD | accuracy (%): 53.0 -> 60.2 | Helped |
Closest papers by meaning
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.91)
- Understanding and Increasing Efficiency of Frank-Wolfe Adversarial Training (CVPR 2022 · similarity 0.90)
- Learning to Learn Single Domain Generalization (CVPR 2020 · similarity 0.89)
- Adversarial Discriminative Domain Adaptation (CVPR 2017 · similarity 0.89)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.88)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.88)
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.