Date/Time
Date(s) - 04/12/2024
4:30 pm - 5:30 pm

Location: BSB B155

Speaker: Hirad Daneshvar, PhD Candidate, TMU (Supervisor: Dr. Reza Samavi)

Title: Trustworthy Graph Neural Networks

Abstract: Accurate prediction of youth mental health readmissions is essential for timely interventions and improved patient outcomes. Graph Neural Networks (GNNs) offer a promising approach to model complex patient data, but their trustworthiness, especially in high-stakes applications like healthcare, is paramount. My research addresses this challenge by introducing a GNN-based model that leverages graph embedding to capture intricate patient visits to the emergency department. The confidence reported by GNN models in their predictions is an important aspect of trustworthiness. We introduce a loss function to calibrate GNN predictions during training. The evaluation results demonstrate the effectiveness of GNNs in capturing the complex semantics available in the patient data as well as the effectiveness of the proposed calibration technique.