Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness and Generalization via Generative Adversarial Training
Omid Poursaeed, Tianxing Jiang, Harry Yang, Serge Belongie, Ser-Nam Lim
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 |
|---|---|---|---|
| adversarial training with generative (GAT) unrestricted examples | Inception-v3 · LSUN clean test | accuracy: 88.9% -> 89.5% | Improved |
| adversarial training with GAT examples | Inception-v3 · LSUN adversarial test | accuracy: 0.0% -> 78.4% | Improved |
| adversarial training with GAT examples | gender classifier · CelebA-HQ clean test | accuracy: 95.7% -> 96.2% | Improved |
| adversarial training with GAT examples | DeepLab-v2 · segmentation clean test | accuracy: 67.9% -> 69.1% | Improved |
| adversarial training with GAT examples | RetinaNet · detection clean test | average precision: 39.0% -> 40.2% | Improved |
| adversarial training with GAT examples | cars vs cats classifier · ImageNet out-of-domain | accuracy: 91.4% -> 92.0% | Improved |
| adversarial training with GAT examples | cars vs cats classifier · LSUN in-domain | accuracy: 94.2% -> 94.7% | Improved |
| train with GAT instead of AT PGD, mean over unseen attacks | Inception-v3 · LSUN unseen attacks | mean accuracy: 13.0% -> 42.1% | Improved |
| train with GAT instead of AT PGD on PGD attack | Inception-v3 · LSUN PGD attack | accuracy: 56.7% -> 39.4% | Hurt |
| train with GAT instead of PAT, mean over unseen attacks | Inception-v3 · LSUN unseen attacks | mean accuracy: 36.5% -> 42.1% | Improved |
| train with GAT instead of PAT, clean accuracy | Inception-v3 · LSUN clean | accuracy: 72.4% -> 89.5% | Improved |
| replace GAT examples with norm-bounded IFGSM-2 examples (ablation) | Inception-v3 · LSUN clean | accuracy: 89.5% -> 88.4% | Hurt |
Closest papers by meaning
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.90)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.90)
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.90)
- Exploring the Relationship Between Architectural Design and Adversarially Robust Generalization (CVPR 2023 · similarity 0.90)
- SGAN: An Alternative Training of Generative Adversarial Networks (CVPR 2018 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
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.
