Our project is supported by the CCF-Huawei Populus Grove Fund (CCF-华为胡杨林基金数据库专项). The goal of this project is to focus on Tabular Data Governance for AI Tasks. The core challenge is: LLMs have strong generalization capabilities, while databases contain vast amounts of accurate knowledge and excel in handling tabular data. Can we combine the strengths of both to help AI models handle tasks that involve tabular data?
Figure1: Venue of the Academic Report Presentation
In August 2024, we presented the kickoff report in Urumqi, Xinjiang [1], where Dr. Wang Zheng shared our approach and gave a talk titled “Tabular Data Governance for AI Tasks.” The main points of the talk included:
- Tabular data is fundamental for AI tasks: The clear, structured format of tabular data makes it easy for computers to process and analyze. It is widely used in key areas like artificial intelligence, data analysis, and business decision-making because of its efficiency and accuracy in handling complex tasks.
- The growing importance of deep learning for tabular data: In the era of large models, there is a rising focus on deep learning techniques for tabular data. Specifically, relational table learning is emerging as a highly promising research area. This is particularly significant given that relational databases account for 73% of the global data management market, representing a massive opportunity for innovation.
- A scalable solution for tabular data governance: To solve the challenges of Tabular Data Governance, we’ve developed a scalable framework based on the BRIDGE algorithm for relational table learning. This framework, built on the rLLM platform [2], offers a robust approach to integrating AI with relational databases to improve the handling of tabular data across various domains.
Figure2: Overview of the Technical Approach
You can watch the full video of Dr. Wang Zheng’s presentation (in Chinese) on Huawei’s official account on Bilibili:📺 Watch the Video on Bilibili.
References:
[1] CCF-Huawei Populus Grove Fund Kickoff Report https://mp.weixin.qq.com/s/3i8w71V3_4PNDktn-Q_jNg.
[2] Li, W., Huang, X., Zheng, J., Wang, Z., Wang, C., Pan, L., & Li, J. (2024). rLLM: Relational Table Learning with LLMs. arXiv preprint arXiv:2407.20157.