Date: 05/28/2026

Time: 9:00 AM to 11:00 AM EDT

Location: TSRB room 523A

Virtual link: https://teams.microsoft.com/meet/286236651306243?p=GJNt6ZfWFP7VluiuT3

Moamen Soliman

Machine Learning PhD Student

School of Electrical and Computer Engineering

Georgia Institute of Technology

Committee:

Dr. Omer T. Inan (Advisor) — School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Rishikesan Kamaleswaran (Co-advisor) — Department of Surgery, School of Medicine, Duke University

Dr. Thomas Ploetz — School of Interactive Computing, Georgia Institute of Technology

Dr. David Anderson — School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Gilles Clermont — School of Medicine, University of Pittsburgh

Dr. Craig S. Jabaley — School of Medicine, Emory University

Abstract

This dissertation advances machine learning for critical care across sensing, prediction, treatment, and faithful explanation. A wearable chest patch with multimodal demodulation algorithms recovers respiratory mechanics from cardiomechanical signals without per-subject calibration. Building on this, septic shock onset is predicted from continuous physiological waveforms, first through a parsimonious bedside model and then through a multimodal fusion framework externally validated across two academic medical centers. Moving beyond prediction, XHemo pairs an offline reinforcement learning policy for fluid and vasopressor management with attribution-grounded large language model explanations, separating decision fidelity from explanatory faithfulness by anchoring generated reasoning to the same Integrated Gradients evidence the policy used. Together, these aims trace a continuum from measuring patient state to anticipating deterioration, recommending action, and explaining recommendations with traceable evidence, contributing methods toward more trustworthy clinical decision support for critical care.