Research Seminar by Chang-Tien Lu | AI for Anticipating Urban Risks: Enhancing Social Stability and Public Health

AI for Anticipating Urban Risks: Enhancing Social Stability and Public Health

Urban environments face increasing risks driven by rapid social, economic, and public health dynamics, many of which are first signaled through social media and other open data sources. This talk examines how AI-powered systems can transform these digital signals into early warnings that support proactive urban risk management, social stability, and public health preparedness. I introduce EMBERS, a real-time forecasting system that anticipates emerging civil unrest by modeling spatiotemporal patterns in open-source data streams using techniques such as Dynamic Query Expansion (DQE) and time-sensitive event modeling; deployments across multiple Latin American countries demonstrate how early indicators of social instability can be systematically extracted from noisy, unstructured online discourse to inform timely decision-making. I then present SimNest, a deep learning framework that integrates social media signals with epidemiological modeling to enable real-time influenza risk assessment at both city and regional scales, where a multi-task spatiotemporal learning approach addresses challenges related to data sparsity, geographic heterogeneity, and real-time prediction. Together, these systems illustrate how AI and social data can be used to anticipate urban risks before they escalate, shifting crisis management from reactive response to proactive mitigation and contributing to more resilient, stable, and healthier cities.