https://peertube.laas.fr/videos/watch/4244417f-9b0e-421b-ad85-06c895894ab8 Talk by Samuel Gruffaz (TUNI): Monday, May 18th, 2026 -- 11.00 - 12.00 -- Room T2 Computer Science Building Riemannian Metric Learning: an Eldorado for AI?  by Samuel Gruffaz, Univ. Tampere (https://sites.google.com/view/samuelgruffaz) Abstract : Riemannian metric learning has recently emerged across several research communities, ranging from generative modeling to causal inference. The growing interest in this topic stems from the fact that a Riemannian metric provides a powerful modeling tool: it defines the underlying geometry of a space, enabling meaningful generalizations of straight lines through geodesics and offering a principled way to measure distances, variability, and deformation. Choosing an appropriate metric is therefore a critical modeling decision, with significant consequences for both interpretation and performance. This problem has been formally and systematically addressed in a recent review, which provides a unifying framework and clarifies key theoretical and practical challenges. In this presentation, we summarize the main contributions of this review and illustrate them through selected examples. Selected reference: Riemannian Metric Learning: Closer to You than You Imagine Selected reference: Riemannian Metric Learning: Closer to You than You Imagine Bio: Samuel Gruffaz is currently a postdoctoral researcher at Tampere University in the Biostatistics group. His research lies at the intersection of Bayesian computation, stochastic optimization, geometric machine learning, and time series analysis, with applications to biomedical data. His recent work focuses in particular on MCMC-based stochastic approximation methods, including SAEM with approximate samplers such as ULA and MALA, as well as geometric and interpretable representations of physiological time series using time warping, dictionary learning, and Riemannian methods. More broadly, his research explores how sampling algorithms, geometry, and latent-variable structure affect the reliability and scalability of modern Bayesian inference procedures. (edited) Thu, 21 May 2026 00:26:46 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://peertube.laas.fr https://peertube.laas.fr/lazy-static/avatars/1921a1bb-9efc-4677-ac93-a455a5a2e107.png https://peertube.laas.fr/videos/watch/4244417f-9b0e-421b-ad85-06c895894ab8 All rights reserved, unless otherwise specified in the terms specified at https://peertube.laas.fr/about and potential licenses granted by each content's rightholder.