Tabular GANs for uneven distribution

TL;DR

This paper surveys three paradigms for tabular data generation—GANs, diffusion models, and LLMs—and introduces a unified framework with a reference implementation.

Contribution

It presents a modular, unified framework supporting multiple generative paradigms for tabular data and validates its effectiveness through experiments on benchmark datasets.

Findings

GAN-based augmentation improves downstream performance under distribution shift

The framework supports data preprocessing, training, inference, and evaluation for all three paradigms

The reference implementation is publicly available for reproducibility

Abstract

Generative models for tabular data have evolved rapidly beyond Generative Adversarial Networks (GANs). While GANs pioneered synthetic tabular data generation, recent advances in diffusion models and large language models (LLMs) have opened new paradigms with complementary strengths in sample quality, privacy, and controllability. In this paper, we survey the landscape of tabular data generation across three major paradigms - GANs, diffusion models, and LLMs - and introduce a unified, modular framework that supports all three. The framework encompasses data preprocessing, a model-agnostic interface layer, standardized training and inference pipelines, and a comprehensive evaluation module. We validate the framework through experiments on seven benchmark datasets, demonstrating that GAN-based augmentation can improve downstream performance under distribution shift. The framework and its reference implementation are publicly available at https://github.com/Diyago/Tabular-data-generation, facilitating reproducibility and extensibility for future research.