More about HKUST
Predictive Routing Query Processing and Movement Analysis in Road Networks
PhD Thesis Proposal Defence Title: "Predictive Routing Query Processing and Movement Analysis in Road Networks" by Miss Jing ZHAO Abstract: Routing query processing in road networks is fundamental for modern transportation applications. It aims to find optimal paths between origins and destinations by considering traffic conditions, network topology, and user preferences. To achieve good query results, we need 1) accurate and efficient traffic condition prediction, 2) corresponding route planning and navigation, and 3) targeted user movement analysis. While recent years have seen advances in route planning, traffic prediction, and trajectory prediction, these components are typically treated separately and face individual challenges. Current route planning methods either assume static costs (based on road segment length or historical travel times) or static cost functions, failing to adapt to real-time traffic evolution and resulting in suboptimal navigation. Existing traffic prediction approaches inefficiently process entire road networks for fixed periods of time, leading to high prediction workload, data throughput, and excessive GPU utilization. Moreover, existing trajectory prediction methods focus on general travel preferences without addressing cold-start users or trajectories. Therefore, we propose an integrated Predictive Routing Query Processing and Movement Analysis framework to address the following critical challenges: 1) the efficiency and scalability of traffic prediction, 2) the maintenance of route optimality under dynamic traffic evolution, and 3) the accuracy of personalized trajectory prediction. Specifically, we propose a just-in-time continuous routing method adaptive to real-time traffic changes, a routing-oriented traffic prediction approach that reduces computational overhead both spatially and temporally, and a discriminative learning method for personalized trajectory prediction. Extensive experiments on real-world road networks demonstrate that our methods achieve enhanced route optimization, improved prediction efficiency, and more accurate personalized travel preference learning compared to existing state-of-the-art methods. Date: Thursday, 16 January 2025 Time: 3:00pm - 5:00pm Venue: Room 5506 Lifts 25/26 Committee Members: Prof. Xiaofang Zhou (Supervisor) Prof. Ke Yi (Chairperson) Prof. Qiong Luo Dr. Lei Li