Research map / Adversarial robustness and security / Certified and empirical robustness
Wasserstein distributional robustness of neural networks
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 |
|---|---|---|---|
| Use ReDLR instead of CE in W-PGD | W-PGD on RobustBench l∞ networks · CIFAR-10, W₂ threat, δ=8/255 | adversarial accuracy: 61.32% -> 45.46% | Hurt |
| Use ReDLR instead of DLR in W-PGD | W-PGD on RobustBench l∞ networks · CIFAR-10, W₂ threat, δ=8/255 | adversarial accuracy: 79.00% -> 45.46% | Hurt |
| Use ReDLR instead of CE in W-PGD | W-PGD on RobustBench l₂ networks · CIFAR-10, W₂ threat, δ=1/2 | adversarial accuracy: 74.62% -> 61.69% | Hurt |
| Use ReDLR instead of DLR in W-PGD | W-PGD on RobustBench l₂ networks · CIFAR-10, W₂ threat, δ=1/2 | adversarial accuracy: 78.69% -> 61.69% | Hurt |
| Use a distributional W₂ threat instead of pointwise W∞ | RobustBench l∞ networks with ReDLR attack · CIFAR-10, δ=8/255 | adversarial accuracy: 57.66% -> 45.46% | Hurt |
| Use a distributional W₂ threat instead of pointwise W∞ | RobustBench l₂ networks with ReDLR attack · CIFAR-10, δ=1/2 | adversarial accuracy: 75.78% -> 61.69% | Hurt |
| Estimate robustness with the proposed Rᶫ and Rᵘ bounds instead of AutoAttack | CIFAR-10 models on RobustBench · CIFAR-10, W∞ threat, δ=8/255, batch size 100 | computation time (seconds): 197 -> 0.52; 175 -> 0.49; 271 -> 0.17; 401 -> 0.55; 456 -> 0.53; 2369 -> 1.46 | Helped |
Closest papers by meaning
- A Unified Wasserstein Distributional Robustness Framework for Adversarial Training (ICLR 2022 · similarity 0.94)
- Certifying Some Distributional Robustness with Principled Adversarial Training (ICLR 2018 · similarity 0.90)
- Distributional Robustness with IPMs and links to Regularization and GANs (NeurIPS 2020 · similarity 0.90)
- Wasserstein Adversarial Examples via Projected Sinkhorn Iterations (ICML 2019 · similarity 0.90)
- Provable Robust Overfitting Mitigation in Wasserstein Distributionally Robust Optimization (ICLR 2025 · similarity 0.89)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 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.