Research map / Image generation / Fast diffusion sampling
Formulating Discrete Probability Flow Through Optimal Transport
Pengze Zhang, Hubery Yin, Chen Li, Xiaohua Xie
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace SDDM reverse generator with discrete probability flow generator (DPF) | SDDM · Synthetic 2spirals, dim 32 state 2 | CSD: 14.3053 -> 2.1719 | Improved |
| use DPF generator instead of SDDM | SDDM · Synthetic 8gaussians, dim 32 state 2 | CSD: 14.1882 -> 1.7945 | Improved |
| use DPF generator instead of SDDM | SDDM · Synthetic checkerboard, dim 32 state 2 | CSD: 14.7433 -> 2.0693 | Improved |
| use DPF generator instead of SDDM, state size 5 | SDDM · Synthetic 2spirals, dim 16 state 5 | CSD: 14.4645 -> 1.9711 | Improved |
| use DPF generator instead of SDDM, state size 5 | SDDM · Synthetic checkerboard, dim 16 state 5 | CSD: 14.6963 -> 1.4172 | Improved |
| use DPF generator instead of SDDM, state size 10 | SDDM · Synthetic 2spirals, dim 12 state 10 | CSD: 12.8463 -> 1.8123 | Improved |
| use DPF generator instead of SDDM, state size 10 | SDDM · Synthetic checkerboard, dim 12 state 10 | CSD: 13.0158 -> 1.1348 | Improved |
| use DPF generator instead of SDDM | SDDM · Synthetic 2spirals, dim 32 state 2 | MMD (lower is better): 2.18e-06 -> 1.89e-05 | Hurt |
| use DPF generator instead of SDDM | SDDM · Synthetic circles, dim 32 state 2 | MMD (lower is better): 6.22e-06 -> 3.27e-05 | Hurt |
| use DPF generator instead of SDDM, state size 5 | SDDM · Synthetic 8gaussians, dim 16 state 5 | MMD (lower is better): 1.01e-4 -> 5.87e-4 | Hurt |
| use DPF generator instead of SDDM, state size 10 | SDDM · Synthetic 2spirals, dim 12 state 10 | MMD (lower is better): 5.52e-4 -> 7.19e-4 | Hurt |
| use DPF generator instead of SDDM | SDDM · Synthetic data, all three state sizes | sample quality (visual) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SDDM — SDDM: Score-Decomposed Diffusion Models on Manifolds for Unpaired Image-to-Image Translation (ICML 2023) · better on 11 of 20 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.86)
- Convergence of Score-Based Discrete Diffusion Models: A Discrete-Time Analysis (ICLR 2025 · similarity 0.86)
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.86)
- Masked Diffusion Models as Energy Minimization (NeurIPS 2025 · similarity 0.85)
- How Discrete and Continuous Diffusion Meet: Comprehensive Analysis of Discrete Diffusion Models via a Stochastic Integral Framework (ICLR 2025 · similarity 0.85)
- NETS: A Non-equilibrium Transport Sampler (ICML 2025 · similarity 0.85)
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.
