The School of EECS is hosting the following HDR Progress Review Confirmation Seminar:

Speaker: Fidan Karimova

Host/Chair: Prof Gianluca Demartini

Urban event prediction remains a critical yet underdeveloped area in spatiotemporal machine learning, particularly when it comes to generalizability, contextual relevance, and adaptability. My PhD research addresses three major gaps: (1) the lack of cross-city generalization due to inconsistent event categorizations and heterogeneous dynamics; (2) the absence of contextual awareness in existing models, which ignore societal signals such as policy shifts or public events; and (3) the inability of current methods to adapt to rapid, non-stationary changes in urban patterns. To tackle these challenges, I first developed HYSTL, a hypernetwork-based framework that learns from multiple cities with mismatched event types by leveraging a structured knowledge graph to guide prediction. Building on this foundation, future work will incorporate real-time societal signals (e.g., news, policies) to improve interpretability and predictive responsiveness, and develop lightweight, adaptive models that maintain performance under dynamic conditions without costly retraining. Collectively, this research advances the field toward more robust, transferable, and context-aware event prediction systems.

Fidan Karimova is a Ph.D. candidate at the University of Queensland, Australia. Prior to her candidacy, she received her bachelor’s degree in Computer Science and master’s degree in Data Science and Artificial Intelligence from University of Strasbourg. She is currently working on the research topic of spatial-temporal predictive modeling and graph representation learning, advised by Dr. Rocky (Tong)  Chen and Prof. Shazia Sadiq.