CDMD: A Cross-Dataset Mixed-Type Diffusion Model for Tabular Data
CDMD trains one mixed-type diffusion model jointly across heterogeneous tabular schemas and beats single-dataset baselines.
CDMD is a tabular diffusion model trained jointly across datasets with different schemas and mixed numerical and categorical features. A schema-restricted reverse process and a shared schema-aware Transformer denoiser handle varying vocabularies end to end. On seven real-world datasets, one jointly trained model achieves the highest average generation quality among single-dataset and cross-dataset baselines while using substantially fewer total parameters. Pretraining on 337 datasets improves results on unseen datasets under limited target data and few adaptation epochs.
- Diffusion is defined directly on mixed numerical and categorical features.
- Schema-restricted reverse process adapts each categorical output to its vocabulary.
- One joint model leads average quality on seven datasets with fewer total parameters.
- Pretraining on 337 datasets improves generation for previously unseen tables.
From the original article
1 min read at arxiv.orgGenerative models for tabular data are typically trained separately for each dataset, limiting knowledge transfer and requiring the storage of many specialized models. In this paper, we introduce CDMD, a tabular diffusion model trained jointly across heterogeneous datasets with different schemas and variable numbers of numerical and categorical features.
By Mohamed Amine Ketata · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.