Title: Spatio-Temporal Data Mining: From Deep Learning to Foundation Models
Abstract: Spatio-temporal (ST) data, which integrates spatial (location-based) and temporal (time-based) information, is the key to improving many real-world applications with high social impacts, such as transportation, logistics, climate, and energy. This talk presents the evolution of spatio-temporal AI, transiting from task-specific deep learning models to the emerging paradigm of foundation models. I will highlight my research progress in modeling human mobility behavior and logistics, specifically focusing on methodologies such as DeepRoute , Graph2Route , and DRL4Route for route prediction and DiffSTG for probabilistic graph forecasting. Furthermore, I will introduce LaDe, the first comprehensive last-mile delivery dataset from industry. Of particular focus in this talk is the recent progress in Spatio-Temporal Foundation Models (STFMs), including the Taxonomy of current STFM and future research opportunities.
Bio: Haomin Wen is postdoctoral researcher at CMU Data Analytics Techniques Algorithms (DATA) Lab. He was a joint PhD at National University of Singapore, received his Ph.D. from INSIS Lab, Beijing Jiaotong university. Dr. Wen’s research mainly focuses on spatial-temporal data mining, human mobility learning to understand the patterns of human’s movements especially in the logistics system. His articles have been published in top journals and conferences, including TKDE, TITS, KDD, ACM SIGSPATIAL, AAAI, etc. He is the founder of Overleaf Copilot. He serves as a reviewer or for the program committee for various journals and conferences, including KDD, IJCAI, ICML, etc.
Personal website: https://wenhaomin.github.io/
