March 4, 2026 at 1:30 PM
Artificial Intelligence promises transformative advances in healthcare, yet real-world medical data challenges nearly every assumption underlying standard machine learning. The healthcare AI revolution is slowed by fundamental constraints: datasets are often high-dimensional, incomplete, privacy-restricted, socially biased, and sometimes extremely scarce, particularly in the context of rare diseases. These characteristics make model development, validation, and deployment far more complex than benchmark performance metrics suggest. In this seminar, we examine how AI systems can be rigorously designed and evaluated under such constraints, addressing privacy-preserving learning, hypothesis validation, structured and non-random missing data, fairness across demographic groups, and robust biomarker selection in high-dimensional settings. Moving beyond predictive accuracy alone, we will critically explore reliability, interpretability, and ethical responsibility as central pillars for building trustworthy machine learning systems in healthcare.
Zaineb Garcia, équipe ORKAD / Endomic
AI Master DS Seminar
ESPRIT Building, Room Agora (RDC)
français
English