Where: Nancy — Faculté des Sciences et Technologies, Université de Lorraine, Boulevard des Aiguillettes, Vandoeuvre-lès-Nancy.
When: Wednesday June 10th 2026
Registration: free but mandatory (before May 27) (registration link) You are welcome to propose a short talk or poster presentation.
This one-day workshop will focus on the theoretical study of neural networks and tensor decompositions using geometric tools. The main topic is the geometry of the corresponding algebraic varieties: neurovarieties (in case of neural networks) and secant varieties (for tensor decompositions). In machine learning theory, understanding geometry of neurovarieties has proven to be the key to reveal many of their fundamental properties such as their identifiability, expressivity, and the behavior of optimization algorithms (see, for example, neuroalgebraicgeometry.ai ). The workhop will present recent developments and discuss connections between neural networks and tensor decompositions. This is a follow-up of the workshop on geometry of tensors organized in 2025.
The topics of the workshop include, but not limited to:
- geometry of low-rank matrix/tensor decompositions
- geometry of neural networks
- neurovarieties and secant varieties, X-rank decompositions
- uniqueness/identifiability of models
- expressivity
- optimization and characterization of critical points
Webpage: https://cran-simul.github.io/geometry-nnets-tensors-2026/
