PhD position: Multicentre pre-training of ICU foundation models
Medical University of Innsbruck | 3–4 years | fully funded | start as soon as possible (flexible)

Project: GPT-MEDIC (ERC Starting Grant)

Host: Institute of Clinical Epidemiology, Public Health, Health Economics,

Medical Statistics and Informatics

Supervisor: Patrick Rockenschaub

In this ERC-funded PhD, you will develop methods for pre-training on ICU event streams across multiple hospitals. The core challenge is building foundation models that generalise across sites without learning sitespecific
shortcuts or memorising sensitive information. ICU data are not like text or images. Events are irregular, values continuous, and measurements depend on clinical decisions that vary by hospital, clinician, and patient state. These quirks make generalisation difficult and privacy risks real, which is why your work will explore multicentre and potentially federated approaches to
pre-training. Federated learning matters because ICU data often cannot be pooled. The goal is foundation models that transfer across settings and support many clinical outcomes. You will have access to ~1M ICU patients and ~33B events, with around half available from day one. The project involves close collaboration with Amsterdam UMC, UCL, and Cambridge.

for more details: PhD Position