Program
TBA
Keynote Speakers
Xiaofang Zhou
Otto Poon Professor of Engineering and Chair Professor
Head of Department, The Hong Kong University of Science and Technology
Title: Bridging the Semantic Gap: Efficient Graph-based Query Processing
Abstract: When users query databases, a fundamental "semantic gap" exists: the user's intent often differs lexically from how the data is stored. While traditional database research has explored fuzzy querying, advances in Large Language Models (LLMs) and specialized vector embedding models provide a modern, highly accurate mechanism for measuring semantic similarity. This talk explores why graph databases are the ideal medium for these semantic queries, offering schema freedom and optimized multi-hop relationship traversal. We will introduce a robust database query optimization foundation based on Conjunctive Regular Path Queries (CRPQ). Building upon this framework, we demonstrate how to implement a "semantic layer" by introducing new semantic operators, similarity metrics, and matching algorithms. Finally, we will showcase how extending CRPQ to semantic matching directly empowers downstream applications like Knowledge Base Question Answering (KBQA), Graph RAG, and Agent Memory systems, while outlining future directions in hypergraph querying.
Bio: Professor Xiaofang Zhou is Otto Poon Chair Professor of Computer Science and Engineering at HKUST and Head of the Department of Computer Science and Engineering. His work covers database systems, data quality management, big data analytics, machine learning and AI. He served as Program Committee Chair for IEEE ICDE 2013, ACM CIKM 2016, and PVLDB 2020, and was General Chair for ICDE 2025 and ACM Multimedia 2015. Before joining HKUST in 2021, he was a Professor of Computer Science and Head of the Data Science discipline at the University of Queensland. He is a Global STEM Professor of Hong Kong and a Fellow of IEEE.
Helen Huang
Professor and Discipline Leader, Data Science, The University of Queensland
Title: Learning at Test Time: Adapting Models to a Changing World
Abstract: AI models are increasingly deployed in dynamic environments that differ from their training conditions. Test-time adaptation addresses this challenge by enabling models to adjust using information available only after deployment. This talk will present recent advances in test-time learning, beginning with computer vision and 3D perception under domain shifts and corruptions, and then extending to graph learning. I will discuss our recent work on continual adaptation for LiDAR-based 3D object detection, robust test-time node classification, and test-time prompt tuning for graph foundation models. These studies highlight a broader view of test-time learning: adaptation can involve model updates, model reuse, structural transformation, or prompt tuning. I will conclude with key challenges and opportunities for building AI models that continue to adapt in changing environments.
Bio: Helen Huang is a Professor and the Discipline Leader for Data Science in the School of EECS at The University of Queensland. Her research interests mainly include multimedia indexing and search, computer vision, recommender systems, and knowledge discovery. Helen has served as an Associate Editor for The VLDB Journal, ACM Computing Surveys, ACM Transactions on Information Systems (TOIS), IEEE Transactions on Multimedia (TMM), Pattern Recognition Journal, etc. and also a member of the VLDB Endowment Board of Trustees. Helen has been elevated to IEEE Fellow and ACM Fellow for contributions to large-scale multimedia content understanding, indexing and retrieval.