聲明:本文來(lái)自于（ID:ykqsd.com）授權(quán)轉(zhuǎn)載發(fā)布。

reported by IT之家, March 11, a joint innovation project on database AI has been launched by Tsinghua University, Chinese tech company Tencent Cloud, and Intel. On March 7, the three organizations announced the start of the "Database AI Joint Innovation Plan," focusing on cross-technology research between databases and large language models to advance database intelligence and establish efficient data infrastructure for AI applications. This marks the first academic-industry collaboration project in China dedicated to database AI research.

With the rise of large language models, traditional databases are facing the challenge of transitioning from "passive storage" to "active services." Conventional databases rely on manual query construction, which is difficult for non-technical users. In contrast, AI models offer a user-friendly interface with minimal technical requirements. At the same time, the demand for massive data processing during AI training and inference necessitates significant improvements in database performance, particularly in distributed architecture and caching management.

This joint innovation project outlines two key technical paths to address these challenges:

In the Text-to-SQL direction, the project team has developed an end-to-end technical framework to tackle key issues such as semantic understanding, pattern matching, and SQL validation. By leveraging techniques like model fine-tuning, reinforcement learning (DPO), and retrieval enhancement (RAG), the system can decompose user queries into subtasks and dynamically match table structures, ultimately generating highly accurate SQL statements.

To accelerate AI inference, the project team has designed a distributed caching system that implements two-stage task scheduling, KVCache optimization, and GPU direct storage technologies. These innovations significantly reduce resource wastage during computation.

Tsinghua University's database research group emphasized, "In the future, databases will not only serve as data storage tools but also become integral components of the AI ecosystem, acting as data warehouses. We look forward to contributing Chinese solutions to the global database technology landscape through this collaboration."