User talk: Cohort amortized personalization: navigating privacy-utility trade-offs in digital twin infrastructure
This talk introduces Cohort-Amortized Personalization (CAP), a framework that addresses the tension between personalized virtual brain twin modeling and international data privacy regulations. CAP learns mappings from shareable cohort statistics to model parameters, enabling anonymous training on public infrastructure while maintaining personalized inference. Through two operational models—demonstrated with aging and epilepsy cohorts—we show how CAP navigates the privacy-utility trade-off compared to federated learning and synthetic data approaches, providing a practical pathway for scalable, GDPR-compliant neuroimaging infrastructure.
Moderator: Ausra Saudargiene
Who You’ll Be Hearing From
This session brings together expert voices from across the EBRAINS community and beyond. Discover the people sharing their insights, research, and perspectives on the topic.
Marmaduke Woodman is a research engineer at the Institute of Systems Neuroscience (INS) at Aix Marseille University, where he develops computational infrastructure for virtual brain twin modeling. His work focuses on bridging mechanistic whole-brain models with clinical applications through scalable inference methods and privacy-preserving architectures. He is a core developer of The Virtual Brain software and contributes to EU projects including EBRAINS 2.0 and Virtual Brain Twin, addressing challenges at the intersection of computational neuroscience, neuroinformatics, and data governance.
Find your way on the map
