Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Lipschitz Regularization
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 ALP instead of GP in Progressive GAN | Progressive GAN · CelebA-HQ | FID: 8.69 -> 14.65 | Hurt |
| Use adversarially selected input pairs for explicit Lipschitz penalties | WGAN critic · CIFAR-10 | FID: 18.86 ± .13 -> 12.96 ± .35 | Helped |
| Replace VAT's KL-based output distance with logit-space mean squared difference and randomize epsilon (ablation) | ConvLarge semi-supervised classifier · CIFAR-10 test | accuracy: 82.46% -> 83.54% | Helped |
| Replace VAT's KL-based output distance with logit-space mean squared difference and randomize epsilon (ablation) | ConvLarge semi-supervised classifier · CIFAR-10 validation | accuracy: 84.2% -> 85.3% | Helped |
| Add batch normalization to the critic and use ALP (ablation) | WGAN-ALP · CIFAR-10 | Inception Score (best during training): 8.34 ± .06 -> 8.71 | Helped |
| Add batch normalization to the critic and use GP (ablation) | WGAN-GP · CIFAR-10 | Inception Score (best during training): 7.86 ± .07 -> 6.29 | Hurt |
| Use a two-sided rather than one-sided explicit Lipschitz penalty (ablation) | WGAN-ALP · CIFAR-10 | training stability | Mixed |
| Vary ξ (ablation) | ALR · Tasks considered | performance | No clear effect |
| Use ALR rather than LP with a critic containing batch normalization (ablation) | Lipschitz-regularized critic · Toy function approximation | gradient smoothness | Helped |
| Use spectral normalization rather than no regularization (ablation) | MLP function approximator · Toy function approximation | gradient match to the optimal mapping | Mixed |
Closest papers by meaning
- On the regularization of Wasserstein GANs (ICLR 2018 · similarity 0.95)
- Lipschitz Generative Adversarial Nets (ICML 2019 · similarity 0.93)
- Improved Training of Wasserstein GANs (NeurIPS 2017 · similarity 0.90)
- Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect (ICLR 2018 · similarity 0.90)
- Towards Generalized Implementation of Wasserstein Distance in GANs (AAAI 2021 · similarity 0.88)
- CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization (CVPR 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.