Research map / Probabilistic and scientific modeling / Variational and generative sampling
Categorical Flow Maps
Daan Roos, Oscar Davis, Floor Eijkelboom, Michael M. Bronstein, Max Welling, İsmail İlkan Ceylan, Luca Ambrogioni, Jan-Willem van de Meent
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 |
|---|---|---|---|
| add simplex-constrained endpoint parametrisation instead of unconstrained velocity (Naive Flow Map) (ablation) | flow map · ZINC (NFE 1) | validity: 59.5 -> 93.5 (ECLD) | Improved |
| add simplex-constrained endpoint parametrisation (CSD) instead of Naive Flow Map (ablation) | flow map · QM9 (NFE 2) | FCD: 0.69 -> 0.49 | Improved |
| add simplex-constrained endpoint parametrisation (ECLD) instead of Naive Flow Map (ablation) | flow map · ZINC (NFE 2) | FCD: 16.1 -> 12.2 | Improved |
| use ECLD instead of CSD | categorical flow map · QM9 (NFE 1) | validity: 90.0 -> 95.8 | Improved |
| use ECLD instead of CSD | categorical flow map · ZINC (NFE 1) | validity: 73.5 -> 93.5 | Improved |
| use CSD instead of ECLD | categorical flow map · QM9 (NFE 1) | FCD: 2.16 -> 2.14 | Improved |
| add flow-map lookahead to guidance instead of denoiser lookahead (ablation) | test-time guidance · binarised MNIST | FID: 38.7 -> 36.9 | Improved |
| add flow-map lookahead to guidance instead of no lookahead (ablation) | test-time guidance · binarised MNIST | FID: 40.7 -> 36.9 | Improved |
| add flow-map lookahead to guidance instead of denoiser lookahead (ablation) | test-time guidance · binarised MNIST | target-class accuracy: 82.5 -> 84.7 | Improved |
| add flow-map lookahead to guidance instead of no lookahead (ablation) | test-time guidance · binarised MNIST | target-class accuracy: 83.1 -> 84.7 | Improved |
| sample with flow map instead of Euler integration of learned velocity (ECLD) | ECLD · QM9/ZINC | FCD | No clear effect |
| add straight-through discretisation in reward gradient instead of soft simplex input | test-time guidance · binarised MNIST | guidance strength | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MoFlow — MoFlow: One-Step Flow Matching for Human Trajectory Forecasting via Implicit Maximum Likelihood Estimation based Distillation (CVPR 2025) · better on 2 of 2 shared results
- PairFlow — PairFlow: Closed-Form Source-Target Coupling for Few-Step Generation in Discrete Flow Models (ICLR 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
- Variational Flow Matching for Graph Generation (NeurIPS 2024 · similarity 0.92)
- Categorical Flow Matching on Statistical Manifolds (NeurIPS 2024 · similarity 0.90)
- How to build a consistency model: Learning flow maps via self-distillation (NeurIPS 2025 · similarity 0.90)
- Rectifying Categorical Flows on Statistical Manifolds for One-Step Generation (NeurIPS 2026 · similarity 0.89)
- Diff2Flow: Training Flow Matching Models via Diffusion Model Alignment (CVPR 2025 · similarity 0.89)
- Test-time Scaling of Diffusions with Flow Maps (NeurIPS 2026 · similarity 0.88)
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.
