SDS Colloquium, Speaker Shenghao Xia

Title: Knowledge-Fused High-Dimensional Spatial-Temporal Data Analytics 

Abstract: Spatial-temporal data, which consist of sequences of observations continuously collected over time and space, are increasingly prevalent in a wide range of application domains, including infrastructure system monitoring, urban surveillance, structural health monitoring, and smart healthcare. Effectively modeling and analyzing such data is essential for system monitoring, anomaly detection, and decision-making. However, three critical challenges hinder accurate modeling: (1) the absence of pre-annotated data for supervised model training, (2) the need for scalable methods to handle high-dimensional spatial-temporal data, and (3) the presence of substantial noise that can obscure underlying patterns. In this colloquium, I will introduce my recent research that addresses these challenges through the development of a knowledge-fused spatial-temporal data analysis methodology. This methodology explicitly models highdimensional spatial-temporal data structures and incorporates domain-specific knowledge to improve estimation accuracy and robustness, while mitigating the impact of noise. I will illustrate the effectiveness of this methodology through multiple case studies, including applications in water distribution system monitoring and surveillance video segmentation.