Deep Learning Models for Mobility Data Mining

点击翻译按钮获取中文摘要

Advances in location acquisition and tracking devices (e.g., GPS technologies, mobile devices, location sensors) have given rise to the generation of enormous mobility data, consisting of trajectories of moving objects, such as people, vehicles, vessels, drones, and more. Similarly, enormous mobility data is generated by IoT devices (e.g., wearables) coming from sensors, meters, and other interconnected devices. Mobility data is a special type of spatiotemporal data - data that contains information about the space/location of moving objects, over time. Extracting and discovering patterns in large spatiotemporal data is critically important to several real-world applications, including understanding human mobility (e.g., pedestrian mobility mining), health care (e.g., epidemics spreading, gait analysis), smart transportation and urban planning (e.g., traffic forecasting and optimization), and location-based services (e.g., recommendations of points of interest). As the volume and complexity of spatiotemporal data has increased rapidly, turning these data into meaningful actions using conventional statistical modeling and data mining methods remains challenging. Recently, with the advances of deep learning techniques, deep learning models such as convolutional neural network (CNN), recurrent neural network (RNN) and graph neural network (GNN) have enjoyed considerable success in various machine learning tasks because of their powerful learning ability. They have been broadly applied in many areas including computer vision, natural language processing, graph data mining, and time series data prediction, which inspires our proposed research to adopt deep learning models for various mobility data mining tasks. The anticipated outcome of our research is the development of the theory and methods of deep learning spatiotemporal models for effectively mining mobility data. Within this context, several aspects of these models will be investigated related to their feasibility, accuracy, robustness, scalability, and interpretability. Best practices will also be developed for training, testing, and validating spatiotemporal models so they can be safely and effectively deployed to address mobility related problems in diverse settings and domains, ranging from intelligent transportation to location-based services and health. Our research program aligns with the Pan-Canadian Artificial Intelligence (AI) Strategy. Conducting world-class machine learning research in data mining, graph mining, big data analytics has the potential to attract the brightest students from around the world, while helping to retain domestic top research talent. Meanwhile, it will contribute to the training of high-quality personnel (HQP), with first-rate analytical and problem-solving skills that have the capacity to stimulate knowledge-based companies, leveraging AI to boost productivity and innovation, foster economic growth and improve the life conditions in Canada and beyond.