Samir Benkhelifa

Interpretability Audits of Graph Neural Diagnostic Models Under Silent Health Information Exchange Dropouts: A Computer Robustness Evaluation

Graph neural networks (GNNs) used for diagnostic prediction in health information exchange (HIE) settings are typically evaluated on complete graphs, yet real exchanges suffer silent dropouts in which entire facilities stop contributing edges without any missingness indicator. th...