Research map / Learning theory and algorithms / Optimal transport
On the Convergence and Robustness of Training GANs with Regularized Optimal Transport
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 a learned latent representation for the cost instead of pixel-space cost (ablation) | SWGAN on MNIST · MNIST | generated image quality | Helped |
| Use Sinkhorn loss instead of regularized OT as the generator objective (ablation) | SWGAN with L1 cost on MNIST · MNIST | generated image quality | Helped |
| Use regularized OT rather than Sinkhorn loss as the generator objective (ablation) | SWGAN on CIFAR-10 · CIFAR-10 | generated image quality | Hurt |
| Use L1 cost rather than Cosine cost (ablation) | Latent Sinkhorn loss SWGAN · CIFAR-10 | inception score convergence | Helped |
| Use Cosine cost rather than L1 cost (ablation) | Latent Sinkhorn loss SWGAN · CIFAR-10 | visual appeal of generated samples | Helped |
| Use SWGAN instead of the compared OT-GAN and Sinkhorn solver methods | GAN image generation · MNIST | generated image quality | Helped |
| Use SWGAN instead of the compared algorithms | GAN image generation · CIFAR-10 | inception score convergence | Helped |
| Use a small regularization weight for the regularized OT objective (ablation) | SWGAN · GAN training | training stability | Hurt |
Closest papers by meaning
- Parallel Optimal Transport GAN (CVPR 2019 · similarity 0.93)
- Hidden Convexity of Wasserstein GANs: Interpretable Generative Models with Closed-Form Solutions (ICLR 2022 · similarity 0.91)
- A Two-Step Computation of the Exact GAN Wasserstein Distance (ICML 2018 · similarity 0.91)
- Discriminator optimal transport (NeurIPS 2019 · similarity 0.91)
- Kantorovich Strikes Back! Wasserstein GANs are not Optimal Transport? (NeurIPS 2022 · similarity 0.91)
- Generative Modeling Using the Sliced Wasserstein Distance (CVPR 2018 · similarity 0.91)
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.