AI+DB: AI and Database Systems
Research vision
We study the integration of artificial intelligence and database systems: AI for databases (AI4DB) and databases for AI applications (DB4AI). Our goal is to build intelligent, efficient, and autonomous systems and provide high-performance data infrastructure for large models, knowledge augmentation, and intelligent decision-making.
Project support
- National Key R&D Program: Efficient Integrated Multi-model Data Processing Models and Optimization
- NSFC Key Project: Fundamental Theories and Key Technologies for Massive Data Management Supporting Technological Innovation
- National Key R&D Program: Big Data Management Systems for High-end Manufacturing
AI4DB: AI for database systems
Intelligent query optimization
We investigate machine learning and large language models for learned query optimization, plan reuse, learning to rank, cardinality estimation, cost modeling, and plan enumeration, improving optimization quality and execution efficiency under complex workloads.
Learning-enhanced database systems
We explore machine learning and deep reinforcement learning within database kernels, including index recommendation, adaptive optimization, autonomous tuning, and performance prediction.
AI-driven database testing and diagnosis
We investigate large language models and intelligent agents for automated testing, bug diagnosis, and failure analysis. Current topics include LLM-assisted testing, automatic discovery of logical bugs, AI-driven fault localization, and autonomous testing frameworks.
DB4AI: Databases for AI applications
Multi-model data management
We study data integration, unified querying, and collaborative analytics in unified multi-model databases, managing relational data, graphs, JSON documents, and vectors together.
Vector databases and similarity search
We investigate high-performance vector indexes, disk-based vector indexes, and large-scale approximate nearest-neighbor search for generative AI and retrieval-augmented generation (RAG).
Selected publications and patents
Bibliographic records below retain their original titles.
- Practical Parameterized Query Optimization via Efficient Plan Reuse and List-wise Ranking（SIGMOD, 2026）
- LEARNT: A Practical Estimator for Cardinality of LIKE Queries with Formal Accuracy Guarantees (CoRR abs/2605.24308, 2026)
- 基于最小化AES的鲁棒计划选择方法.《小型微型计算机系统》2026年4期
- MSP: Learned Query Performance Prediction Using MetaInfo and Structure of Plans (APWeb/WAIM, 2022)
- A Survey on Advancing the DBMS Query Optimizer: Cardinality Estimation, Cost Model, and Plan Enumeration（Data Science and Engineering, 2021）
- 数据库索引调优技术综述.《计算机研究与发展》2024年4期
- An Index Advisor Using Deep Reinforcement Learning（CIKM 2020）
- Learning an Index Advisor with Deep Reinforcement Learning（APWeb-WAIM 2021）
- 面向自动参数调优的动态负载匹配方法.《计算机应用》2021年3期
- Graph-centric Cross-model Data Integration and Analytics in a Unified Multi-model Database（arXiv）
- LSMDiskANN：更新友好型磁盘向量索引框架. 《软件学报》2026年3期