Spatiotemporal learning for urban mobility and traffic data

点击翻译按钮获取中文摘要

With recent advances in information and communications technology, continuous streams of spatiotemporal mobility and traffic data are generated in cities through various sensing technologies, including loop detectors, cameras, RFID, cellphones, floating cars, and crowdsourcing platforms (e.g., Google Waze). The exponentially growing urban big data provides us with unprecedented opportunities to understand urban mobility and transportation systems in a data-driven way. Efficient and reliable modeling of these spatiotemporal data sets can benefit a wide range of intelligent transportation systems (ITS) and urban planning applications, such as travel demand prediction, trip planning, travel time estimation, route planning, ride sharing, transit service scheduling, signal control, and congestion/disruption management. The key to modeling spatiotemporal mobility/traffic data is to characterize the higher-order correlations/dependencies within the data. However, due to the large-scale, high-dimensional, incomplete, nonlinear, non-stationary and heterogeneous nature of emerging spatiotemporal mobility/traffic data, traditional models become insufficient to serve this role. The field is calling for new concepts and tools based on artificial intelligence and machine learning. The long-term goal of this Discovery program is to establish novel statistical learning and innovative computational methods to learn from urban spatiotemporal big data and provide smart transportation applications for the operation and planning of future smart and resilient cities. More specifically, this program consists of four major research objectives for efficient and reliable spatiotemporal learning based on my recent progress on spatiotemporal data analytics: (1) develop advanced tensor learning and deep learning models for spatiotemporal mobility/traffic data, (2) develop scalable/efficient online learning models for large-scale and real-time problems, (3) develop new prediction schemes capturing long-range spatiotemporal dependencies, and (4) adapt the proposed learning frameworks for data heterogeneity and ensure model reliability. This Discovery program will create state-of-the-art tools and knowledge to model high-dimensional spatiotemporal data sets generated from urban systems, and also provide decision-making tools and ITS applications for smart transportation of the future. In a board sense, the methodological pipeline developed through this program is not only transformative in transportation engineering but also valuable to other smart cities applications related to spatiotemporal modeling, such as weather, air quality and epidemic predictions. In addition, this program will promote fundamental interdisciplinary advances and train interdisciplinary HQP through integrating artificial intelligence knowledge with domain expertise in transportation engineering, contributing to the rapid development of artificial intelligence and smart cities in Canada.