TabDiff: a Multi-Modal Diffusion Model for Tabular Data Generation

Figure 1: Visualing the generative process of TabDiff. A high-quality version of this video can be found at tabdiff_demo.mp4

This repository provides the official implementation of the paper "TabDiff: a Multi-Modal Diffusion Model for Tabular Data Generation".

Latest Update

• [2024.10]：Our code is at the final stage of cleaning up. Please check back soon for its release!

Introduction

Figure 2: The high-level schema of TabDiff

TabDiff is a unified diffusion framework designed to model all muti-modal distributions of tabular data in a single model. Its key innovations include:

1. Framing the joint diffusion process in continuous time,
2. A feature-wised learnable diffusion process that offsets the heterogeneity across different feature distributions,
3. Classifier-free guidance conditional generation for missing column value imputation.

The schema of TabDiff is presented in the figure above. For more details, please refer to our paper.