LLM for Tabular Data Generation
Leveraging causal structure for high-fidelity synthetic tabular data
This project investigates how large language models (LLMs) can be leveraged to generate high-fidelity synthetic tabular data, guided by the underlying causal structure of the data.
Key activities:
- Conducted a systematic review of LLM-based techniques for synthetic tabular data generation.
- Reproduced representative state-of-the-art algorithms and evaluated their performance across diverse datasets.
- Investigating novel methods that bridge the gap between LLM generative capabilities and structured data utility, with causal relationships as a guiding principle.
Manuscript: Dynamic Mechanism Fidelity for Nonstationary Tabular Data: Evaluation and Mechanism-Guided Generation — Zhili Wang, Chen Zhao, Yi He. Under review at AAAI 2027.