Data Science Talk Series 

From Cities to Innovation Ecosystems: Graph-Based Learning for Spatiotemporal Forecasting and Societal Impact 

Abstract: The increasing complexity of modern cities and technologies presents pressing challenges for both urban management and future innovation planning. This talk presents recent research on graph neural networks (GNNs) and multimodal machine learning for spatiotemporal event forecasting and technology foresight. It first introduces a location representation learning framework that models urban environments as dynamic graphs, enabling accurate prediction of crime categories by integrating geospatial features of buildings, amenities, and roads. This talk will then highlight a novel adversarial dynamic graph transformer designed for unlimited-range prediction of traffic incident impacts, which leverages road-anchored spatial attention and temporal anomaly detection to capture long-range dependencies in transportation networks. The talk further covers ongoing work, which aims to forecast emerging technology areas and their regional societal and economic impacts through interpretable, multimodal AI pipelines combining patents, publications, and workforce data. Collectively, these efforts advance interpretable and scalable graph-based learning methods for intelligent transportation, public safety, and innovation policy, while emphasizing the broader impacts on workforce development and equitable regional growth.

Bio of the speaker: Kaiqun Fu is an Assistant Professor of Computer Science at Texas Christian University (TCU). His research focuses on spatial data mining, graph neural networks, and machine learning, with broad applications in intelligent transportation systems, urban safety, power systems, and social networks. His research is supported by multiple federal and state grants, such as NSF CISE, NSF TIP, and NASA AIST, with funded projects totaling over $1.7M. His work has been published in leading venues such as AAAI, IJCAI, ACM SIGSPATIAL, IEEE BigData, and ITSC, with contributions spanning traffic incident impact forecasting, interpretable spatiotemporal modeling, and multimodal AI for emerging technology analysis. Dr. Fu is also actively engaged in academic service, serving as a reviewer/PC member for leading journals and conferences, including IEEE TKDE, ACM TKDD, AAAI, IJCAI, and KDD.