ED 512 - InfoMaths (Informatique et Mathématiques)
Publié le 4 novembre 2026 | Mis à jour le 4 novembre 2026
Knowledge Transfer on Manifolds: Aligning Heterogeneous Brain Ageing Cohorts
Brain ageing unfolds along trajectories that differ markedly between reference cohorts collected under harmonized protocols and smaller, more heterogeneous cohorts gathered locally with different scanners, recruitment criteria, or age ranges, so that a model of anatomical change learned on one population rarely transfers safely to another without careful adaptation. Manifold-valued data such as functional, matrix-valued, image, or shape data arise naturally in this setting cortical and hippocampal surfaces, diffusion tensors, connectivity matrices, or metabolite compositions where learning models are formulated in tangent spaces with a Riemannian metric to exploit intrinsic geometry rather than being flattened into an inappropriate Euclidean representation.