Yasmeen George - Data Science
Date: - Location: Eurecom
Abstract: Foundation models provide powerful representations for medical AI, yet their practical deployment remains challenged by data scarcity, domain shift, and unreliable reasoning. In this talk, I will present our work on test-time adaptation for medical images and question answering. I will also introduce a training-free framework for detecting hallucinations in medical multimodal language models that evaluates whether model predictions genuinely depend on image evidence. These studies highlight emerging opportunities for combining foundation models, domain adaptation, and trustworthy AI to create clinically useful medical AI systems. Bio: Yasmeen George is a Senior Lecturer and Deputy Head of Department in Data Science and Artificial Intelligence at Monash University, and a co-founding member of the AIM for Health Lab and the AI in Health Initiative at Monash, where she leads research at the intersection of reliable AI and healthcare. She received her PhD from the University of Melbourne in 2018 (Australia). Her research focuses on developing robust and scalable artificial intelligence (AI) solutions for medical imaging analysis, disease diagnosis and monitoring, and personalised treatment recommendation. Before joining Monash, Yasmeen worked as a Research Scientist with IBM Research Australia focusing on the use of AI for retinal disease diagnosis and patient monitoring. She has a strong technical focus on domain adaptation, federated learning, medical reasoning, data-efficient learning, and generative AI techniques with a focus on improving model reliability and safety in clinical settings.
