Tight Robustness Certificates and Wasserstein Distributional Attacks for Deep Neural Networks
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918320321593344 |
|---|---|
| author | Le, Bach C. Dao, Tung V. Nguyen, Binh T. Chu, Hong T. M. |
| author_facet | Le, Bach C. Dao, Tung V. Nguyen, Binh T. Chu, Hong T. M. |
| contents | Wasserstein distributionally robust optimization (WDRO) provides a framework for adversarial robustness, yet existing methods based on global Lipschitz continuity or strong duality often yield loose upper bounds or require prohibitive computation. We address these limitations with a primal approach and adopt a notion of exact Lipschitz certificates to tighten this upper bound of WDRO. For ReLU networks, we leverage the piecewise-affine structure on activation cells to obtain an exact tractable characterization of the corresponding WDRO problem. We further extend our analysis to modern architectures with smooth activations (e.g., GELU, SiLU), such as Transformers. Additionally, we propose novel Wasserstein Distributional Attacks (WDA, WDA++) that construct candidates for the worst-case distribution. Compared to existing attacks that are restricted to point-wise perturbations, our methods offer greater flexibility in the number and location of attack points. Extensive evaluations demonstrate that our proposed framework achieves competitive robust accuracy against state-of-the-art baselines while offering tighter certificates than existing methods. Our code is available at https://github.com/OLab-Repo/WDA. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_10000 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Tight Robustness Certificates and Wasserstein Distributional Attacks for Deep Neural Networks Le, Bach C. Dao, Tung V. Nguyen, Binh T. Chu, Hong T. M. Machine Learning Optimization and Control Wasserstein distributionally robust optimization (WDRO) provides a framework for adversarial robustness, yet existing methods based on global Lipschitz continuity or strong duality often yield loose upper bounds or require prohibitive computation. We address these limitations with a primal approach and adopt a notion of exact Lipschitz certificates to tighten this upper bound of WDRO. For ReLU networks, we leverage the piecewise-affine structure on activation cells to obtain an exact tractable characterization of the corresponding WDRO problem. We further extend our analysis to modern architectures with smooth activations (e.g., GELU, SiLU), such as Transformers. Additionally, we propose novel Wasserstein Distributional Attacks (WDA, WDA++) that construct candidates for the worst-case distribution. Compared to existing attacks that are restricted to point-wise perturbations, our methods offer greater flexibility in the number and location of attack points. Extensive evaluations demonstrate that our proposed framework achieves competitive robust accuracy against state-of-the-art baselines while offering tighter certificates than existing methods. Our code is available at https://github.com/OLab-Repo/WDA. |
| title | Tight Robustness Certificates and Wasserstein Distributional Attacks for Deep Neural Networks |
| topic | Machine Learning Optimization and Control |
| url | https://arxiv.org/abs/2510.10000 |