In recent years, it has become clear that geometry plays an important role in understanding the properties of neural networks. This issue lies at the heart of the challenges facing artificial intelligence, whether in terms of explaining its performance or its vulnerabilities. The latter is of particular importance in the context of civil aviation and critical systems in general. This event will bring together specialists from various disciplines conducting research at the interface of geometry and machine learning.

• Stéphanie Allassonnière (Université Paris Cité)  - "Taking advantage of geometry in Variational Autoencoders"

• Pierre Baudot (Median technology) -"Information cohomology and higher-order statistical interactions: theory and ML/AI applications"

• Daniel Bennequin (Institut de Mathématiques de Jussieu-Paris Rive Gauche) -"Homotopy semantics in LLMs, from data, learning and architectures"

• Michel Boyom (Institut Montpelliérain Alexander Grothendieck) -"Koszul homological series and Fedosov foliations on statistical manifolds"

• Mathieu Carrière (INRIA Université Côte d'Azur) -"Diffeomorphic interpolation for efficient topological optimization"

• Rita Fioresi (Università di Bologna) -"Manifold Learning via Foliations and Knowledge Transfer"

• Matthieu Jonckheere (LAAS-CNRS) -"Fermat distance: inferring meaningful distances for machine learning tasks"

• Alexey Lazarev (Université de Toulouse) -"Curvature regularization and Riemannian clustering in the latent space of an autoencoder"
  

• Alice Le Brigant (Université Paris 1 Panthéon-Sorbonne) - "The different ways to interpolate between probability distributions in information geometry"

• Mathieu Serrurier  (Université de Toulouse) -"Self-Supervised Learning with 1-Lipschitz Constraints"

• Alice Barbara Tumpach (Université de Lille) - "Infinite-dimensional Geometry and AI"

• Hong Van Le (Czech Academy of Sciences) - "Probabilistic morphisms  and Bayesian supervised learning"

The meeting is organised under the auspices of the framework of the COST project CA21109 – Cartan geometry, Lie, Integrable Systems, quantum group Theories for Applications (CaLISTA).  The organisers are grateful for funding from the EU via COST.