Research map / Image generation / Autoregressive and latent generation
CoDi: Co-evolving Contrastive Diffusion Models for Mixed-type Tabular Synthesis
Chaejeong Lee, Jayoung Kim, Noseong Park
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 |
|---|---|---|---|
| Process discrete variables in a discrete (categorical) diffusion space instead of continuous space (ablation) | discrete-only tabular data · Car, Clave, Nursery, Phishing | F1 score and coverage: e.g. Phishing coverage 0.127 -> 0.644; Car F1 0.891 -> 0.949 | Improved |
| Add contrastive learning with negative sampling (CoDi vs CoDi w/o CL) (ablation) | CoDi · 11 datasets (classification and regression) | F1 (R^2) and coverage: improves on most datasets, but e.g. Heart F1 0.886 -> 0.872 and Seismic coverage 0.380 -> 0.359 | Mixed |
| Negative condition by shuffling the whole discrete variable set jointly (keeping pairs) vs shuffling two random columns (ablation) | contrastive negative sampling · Heart, Faults, Insurance | F1 (R^2) and coverage: Heart F1 0.872 vs 0.857/0.850; Heart coverage 0.949 vs 0.847/0.820 | Improved |
| Co-evolving conditional diffusion (two models conditioned on each other) with STaSy-style baseline comparison | mixed-type tabular synthesis · Binary classification (avg) | Binary F1: 0.4726 vs 0.4559 (STaSy) | Improved |
| Propose CoDi (two diffusion models with contrastive learning) vs STaSy in sampling time | tabular synthesis runtime · Average over datasets | runtime (10K samples): 0.5187 vs 4.6417 s | Improved |
Later papers that compared with this method
2 later accepted papers report CoDi as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- OmniFlow: Any-to-Any Generation with Multi-Modal Rectified Flows (CVPR 2025 · better on 4 of 5 shared results)
- Multimodal Promptable Token Merging for Diffusion Models (AAAI 2025 · better on 0 of 2 shared results)
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.
- STaSy — STaSy: Score-based Tabular data Synthesis (ICLR 2023) · better on 5 of 5 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
- TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation (ICLR 2025 · similarity 0.90)
- Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Space (ICLR 2024 · similarity 0.90)
- Continuous Diffusion for Mixed-Type Tabular Data (ICLR 2025 · similarity 0.89)
- TabNAT: A Continuous-Discrete Joint Generative Framework for Tabular Data (ICML 2025 · similarity 0.89)
- CTSyn: A Foundation Model for Cross Tabular Data Generation (ICLR 2025 · similarity 0.88)
- ClavaDDPM: Multi-relational Data Synthesis with Cluster-guided Diffusion Models (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.
