Speaker: Prof. Jianliang Xu (徐建良 教授)

Time: 14:00-14:50, 14 March 2025 (Friday)  (Beijing time)

Venue: T2-202, UIC

Language: English

Abstract

Retrieval-augmented generation (RAG) has emerged as a powerful technique that combines retrieval-based and generation-based methods to produce highly accurate and contextually relevant responses in large language models (LLMs). This talk delves into the key concept of vector similarity search, a critical component of RAG, which enables efficient and effective retrieval of relevant documents or data points from large datasets. We will introduce a novel approach that utilizes proximity graphs to enhance search performance in vector databases. Furthermore, we will present FedKNN, a system designed to support secure and oblivious similarity searches in federated settings, ensuring data privacy and security while maintaining high performance.

About the Speaker

徐建良，香港浸会大学计算机系讲座教授、系主任，IEEE Fellow，CCF数据库专委会委员。主要研究方向为数据库、区块链、大数据安全及隐私等，已发表250余篇论文，其中大部分发表在CCF推荐的A类期刊和会议，包括SIGMOD、VLDB、ICDE、TKDE等，获WISE 2019最佳论文、CIKM 2020候选最佳论文。目前H-index为64，论文总引用次数超过14000次。其研究获得国家自然科学基金、香港研究资助局、香港创新科技署等机构以及工业界的资助40余项，总研究经费超过4000万元。担任或曾经担任TKDE、TPDS、TBD等一流国际期刊的编委及多个重要国际会议的程序委员会主席，包括PVLDB 2024、PVLDB2026编委及ICDE 2024、ICDE 2025领域主席。领导研发了香港首个新冠病毒风险警示系统BU-Trace，其技术被应用于超过七百万用户的香港政府疫情防控App「安心出行」。更多详情可访问 : https://www.comp.hkbu.edu.hk/~xujl。