Research map / Image generation / GANs
Learning Generative Models across Incomparable Spaces
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 relational Gromov-Wasserstein matching instead of requiring direct cross-space comparisons | GW GAN · Gaussian mixtures across 2D and 3D spaces | mode and relative-distance recovery | Helped |
| Use shortest-path distances on the sample k-nearest-neighbor graph | GW GAN · S-curve manifold | manifold structure recovery | Helped |
| Represent the reference using graph shortest paths and generate 2D Euclidean samples | GW GAN · Weighted graph | neighborhood structure recovery | Helped |
| Add a style adversary targeting bold font | GW GAN on MNIST · MNIST | digit thickness | Helped |
| Remove the learned adversary (ablation) | GW GAN · 2D Gaussian mixtures | mode recovery | No clear effect |
| Add the learned adversary (ablation) | GW GAN · MNIST | distribution recovery | Helped |
| Add an ℓ1 penalty on generated samples (ablation) | GW GAN · Four-Gaussian mixture | centering around the origin | Helped |
| Remove the ℓ1 penalty on generated samples (ablation) | GW GAN · Four-Gaussian mixture | mode recovery | No clear effect |
| Stop updating the adversary after 10000 iterations (ablation) | GW GAN · GW GAN training | generator stability | No clear effect |
| Use Procrustes-based orthogonal regularization rather than the compared weight-based approaches (ablation) | GW GAN adversary · GW GAN training | adversary distortion control | Helped |
| Replace Procrustes-based regularization with layerwise orthogonality constraints (ablation) | GW GAN adversary · GW GAN training | adversary distortion control | Hurt |
| Add total variation regularization | GW GAN image generation · MNIST, fashion-MNIST, and gray-scale CIFAR-10 | natural image appearance | Helped |
Closest papers by meaning
- On Deep Generative Models for Approximation and Estimation of Distributions on Manifolds (NeurIPS 2022 · similarity 0.89)
- Generative Modeling Using the Sliced Wasserstein Distance (CVPR 2018 · similarity 0.89)
- Disconnected Manifold Learning for Generative Adversarial Networks (NeurIPS 2018 · similarity 0.88)
- A Geometric Analysis of Deep Generative Image Models and Its Applications (ICLR 2021 · similarity 0.88)
- Statistically Optimal Generative Modeling with Maximum Deviation from the Empirical Distribution (ICML 2024 · similarity 0.87)
- Distance Marching for Generative Modeling (NeurIPS 2026 · similarity 0.86)
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.