Research map / Probabilistic and scientific modeling / Variational and generative sampling
PairFlow: Closed-Form Source-Target Coupling for Few-Step Generation in Discrete Flow Models
Mingue Park, Jisung Hwang, Seungwoo Yoo, Kyeongmin Yeo, Minhyuk Sung
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 |
|---|---|---|---|
| train UDLM on pairs from closed-form backward inversion (PairFlow) | UDLM · QM9 1-step | valid molecules: 17.5 -> 223.4 | Improved |
| use closed-form paired data instead of random uniform source pairing (ablation) | UDLM · QM9 1-step | valid molecules: 47.1 -> 223.4 | Improved |
| use closed-form paired data instead of random uniform source pairing (ablation) | UDLM · QM9 64-step | valid molecules: 998.6 -> 1000.2 | No clear effect |
| PairFlow preprocessing instead of ReDi finetuning | UDLM · QM9 2-step | valid molecules: 232.4 -> 416.0 | Improved |
| PairFlow preprocessing instead of DCD finetuning | UDLM · QM9 2-step | valid molecules: 530.8 -> 416.0 | Hurt |
| distill from PairFlow base instead of UDLM base with DCD | UDLM + DCD · QM9 2-step | valid molecules: 530.8 -> 685.8 | Improved |
| distill from PairFlow base with ReDi | UDLM + ReDi · ZINC-250k 2-step | valid molecules: 75.9 -> 221.5 | Improved |
| train with PairFlow pairs | UDLM · MNIST-Binary 1-step | FID: about 40.59 vs UDLM; 68.9% improvement | Improved |
| PairFlow instead of ReDi at 2 steps | UDLM · MNIST-Binary 2-step | FID: 10.36 -> 15.61 | Hurt |
| train with PairFlow pairs | UDLM · CIFAR-10 | FID: 33.65 -> 28.07 | Improved |
| apply DCD or ReDi distillation | UDLM · CIFAR-10 | FID / IS | Hurt |
| build pairs with backward inversion rather than forward-velocity sampling (ablation) | PairFlow pair construction · training set coverage | coverage: forward sampling covers about 63% of training set | Improved |
Later papers that compared with this method
1 later accepted paper reports PairFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Categorical Flow Maps (ICML 2026 · better on 1 of 1 shared result)
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
- On Fitting Flow Models with Large Sinkhorn Couplings (NeurIPS 2026 · similarity 0.88)
- Flow Matching with General Discrete Paths: A Kinetic-Optimal Perspective (ICLR 2025 · similarity 0.87)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.87)
- Bernoulli Flow Models: Self-Consistent Generative Modeling for Binary Data (NeurIPS 2026 · similarity 0.87)
- DiverseFlow: Sample-Efficient Diverse Mode Coverage in Flows (CVPR 2025 · similarity 0.87)
- Align Your Flow: Scaling Continuous-Time Flow Map Distillation (NeurIPS 2025 · 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.
