AI for Synthetic Data Generation
We study generative models for heterogeneous tabular data, including relationships among tables, schemas, and temporal changes.
AI for Synthetic Data Generation
Members: 임재성
Methods: Variational AutoEncoder, Generative Models
Tabular data is highly heterogeneous, often containing continuous and discrete variables together. As a result, generative AI models developed primarily for images or text have limitations when applied directly to tabular data, and various models that reflect its characteristics have recently been developed.
Our lab studies tabular data generation using large language models (LLMs). The data to be generated is not limited to a single independent table: we also consider more complex forms, including changes over time, relationships among multiple tables, and the structure of relational databases.
In particular, we study methods that infer and explain the structure and relationships among tables and variables from a given schema, while generating data that satisfies these structural constraints and semantic relationships. Our ultimate goal is to develop generative models that preserve not only individual variable distributions but also relationships among tables and the structural and temporal characteristics of the data.
Funding: Ministry of Science and ICT
Leader: 임재성