Research map / Image generation / Fast diffusion sampling
Cascaded Flow Matching for Heterogeneous Tabular Data with Mixed-Type Features
Markus Mueller, Kathrin Gruber, Dennis Fok
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 |
|---|---|---|---|
| use a cascaded pipeline with distributional tree latents, data-dependent coupling and flow matching instead of joint CDTD | CDTD · 12 tabular datasets, MNAR 10% missing | detection score: 0.518 -> 0.787 | Improved |
| add cascaded factorization (x_cat first, then x_num given x_cat) to CDTD (ablation) | CDTD · 12 tabular datasets | detection score: 0.518 -> 0.694 | Improved |
| add low-resolution latents z to cascaded CDTD without the flow matching changes (ablation) | CDTD + cascade · 12 tabular datasets | detection score: 0.694 -> 0.175 | Hurt |
| replace CDTD high-resolution model with flow matching (independent coupling, linear paths) (ablation) | CDTD + latents z · 12 tabular datasets | detection score: 0.175 -> 0.719 | Improved |
| add data-dependent coupling from z (ablation) | flow matching high-resolution model · 12 tabular datasets | detection score: 0.719 -> 0.786 | Improved |
| add learnable non-linear time schedules (paths) (ablation) | flow matching with data-dependent coupling · 12 tabular datasets | detection score: 0.786 -> 0.787 | No clear effect |
| use distributional tree encoder instead of Gaussian mixture model for z (ablation) | TabCascade · 12 tabular datasets | detection score: 0.479 -> 0.787 | Improved |
| use TabCascade instead of the best baseline | CDTD · 12 tabular datasets | Wasserstein distance (num): 0.009 -> 0.004 | Improved |
| use TabCascade instead of the best baseline | CDTD · 12 tabular datasets | MLE: 0.039 -> 0.027 | Improved |
| use TabCascade instead of CDTD | CDTD · 12 tabular datasets | MIA score: 0.958 -> 0.935 | Hurt |
| use TabCascade instead of CDTD | CDTD · 12 tabular datasets | DCR share: 0.883 -> 0.890 | Hurt |
| use TabCascade instead of CDTD for categorical features | CDTD · 12 tabular datasets | JSD (cat): 0.020 -> 0.018 | No clear effect |
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.
- TabDiff — TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation (ICLR 2025) · better on 1 of 1 shared result
- TabDDPM — TabDDPM: Modelling Tabular Data with Diffusion Models (ICML 2023) · 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
- Exponential Family Variational Flow Matching for Tabular Data Generation (ICML 2025 · similarity 0.91)
- Continuous Diffusion for Mixed-Type Tabular Data (ICLR 2025 · similarity 0.91)
- TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation (ICLR 2025 · similarity 0.90)
- Discrete Flow Matching (NeurIPS 2024 · similarity 0.88)
- Categorical Flow Matching on Statistical Manifolds (NeurIPS 2024 · similarity 0.87)
- Fisher Flow Matching for Generative Modeling over Discrete Data (NeurIPS 2024 · 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.
