Research map / Learning theory and algorithms / Optimal transport
Distributional Sliced-Wasserstein and Applications to Generative Modeling
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 DSW-100 instead of SW-10000 | generative modeling · LSUN | FID: 46 vs over 60 | Helped |
| use DSW instead of SW at similar computational time | generative modeling · CIFAR10 | FID: about 20 units lower | Helped |
| use DSW instead of SW at similar computational time | generative modeling · MNIST | Wasserstein-2: roughly 10% lower | Helped |
| use DSW-10 instead of PRW-5 | generative modeling · MNIST | Wasserstein-2 | Helped |
| use DSW-10 instead of PRW-5 | generative modeling · MNIST | computational time: 30 times faster | Helped |
| use PRW with 50-dimensional subspace instead of DSW-10 | generative modeling · MNIST | Wasserstein-2: about 29% | Helped |
| use PRW with 50-dimensional subspace instead of DSW-10 | generative modeling · MNIST | computational cost: around 40 times slower | Hurt |
| increase the number of gradient steps used to update the direction distribution (ablation) | DSW · MNIST | generative quality | Helped |
| increase the number of gradient steps used to update the direction distribution (ablation) | DSW · MNIST | computational speed | Hurt |
| remove the direction-diversity regularization (λ_C = 0) (ablation) | DSW · MNIST | projection-direction overlap (A_N): A_N gets close to 1 | Hurt |
| increase direction-diversity regularization to λ_C = 1000 (ablation) | DSW · MNIST | projection-direction overlap (A_N): A_N is close to 0.1 | Helped |
| use DGSW instead of uniform-sampling GSW | generative modeling · generative modeling experiments described in the appendices | generative quality | Helped |
Closest papers by meaning
- Markovian Sliced Wasserstein Distances: Beyond Independent Projections (NeurIPS 2023 · similarity 0.93)
- Asymptotic Guarantees for Learning Generative Models with the Sliced-Wasserstein Distance (NeurIPS 2019 · similarity 0.92)
- Run-Sort-ReRun: Escaping Batch Size Limitations in Sliced Wasserstein Generative Models (ICML 2021 · similarity 0.92)
- Sliced Wasserstein with Random-Path Projecting Directions (ICML 2024 · similarity 0.90)
- Tree-Sliced Wasserstein Distance with Nonlinear Projection (ICML 2025 · similarity 0.90)
- Towards Understanding Gradient Dynamics of the Sliced-Wasserstein Distance via Critical Point Analysis (ICML 2025 · similarity 0.89)
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.