MPUGAT : A Novel Framework for Inferring Dynamic Infectious Disease Transmission with Graph Attention

Epidemic modeling is essential for understanding and managing the spread of infectious diseases. However, it often faces challenges related to unidentifiability due to high-dimensional parameters. Therefore, integrating various data sources to infer epidemic parameters is crucial for reliable modeling. We propose MPUGAT, a hybrid framework that combines a multi-patch compartmental model with a spatiotemporal deep learning approach. By leveraging a Graph Attention Network (GAT), MPUGAT effectively captures spatiotemporal infection patterns from diverse time series data to infer a dynamic transmission matrix. Applied to COVID-19 data from South Korea, MPUGAT demonstrates superior performance in estimating the time-varying transmission matrix, aligning well with real-world dynamics. This framework offers a novel approach to integrating easily accessible, low-dimensional, non-epidemic-related data into epidemic modeling, enhancing both inference and interpretability.

MPUGAT : A Novel Framework for Inferring Dynamic Infectious Disease Transmission with Graph Attention

Monday, July 14 at 6:00pm

Minji Lee

"MPUGAT : A Novel Framework for Inferring Dynamic Infectious Disease Transmission with Graph Attention"

#SMB2025 Follow