Cross-Tabular Data Generation via Diffusion Transformers
August 17, 2026
A two-stage framework enables synthetic generation of data across multiple heterogeneous tables by first converting them into standardized statistical tables. A diffusion transformer then learns structural patterns across these statistics to reconstruct diverse, multi-table datasets.
HOW THIS AFFECTS YOU
●
builderYou can use this method to generate complex synthetic datasets when working with multiple relational tables.
●
researcherThe two-stage statistical transformation approach offers a new way to handle tabular heterogeneity.