NeuralFEM Workshop 2026

Strasbourg, 14-18 December 2026

Monday 14 – Friday 18 December 2026

IRMA conference room, University of Strasbourg

Workshop

The week is about the certified approximation of partial differential equations by neural networks: how to make their predictions not only fast, which they already are, but also reliable, which they are not yet. Once trained, a network answers in near real time — what intra-operative guidance demands — but nothing guarantees that the answer is right, and that is what keeps such methods out of critical applications.

The topics are:

  • neural-network methods for nonlinear and non-smooth problems — variational inequalities from contact and plasticity, fractional operators modelling anomalous diffusion — and hybrid schemes combining them with finite elements;
  • a priori error analysis of the network approximation: function spaces suited to it, how many neurons a given accuracy costs, convergence of the training algorithms;
  • a posteriori error estimation and adaptive strategies, carried over from finite elements to steer the refinement of a network;
  • reduced basis methods, as the competing approach against which the networks are measured;
  • applications in biomechanics and biomedicine: tissue compression, stent–artery interaction, anomalous diffusion in biological tissue.

The workshop is open to anyone working on these questions, and not only to the members of the NeuralFEM project — a MATH-AmSud collaboration between Uruguay, Colombia, Chile and France, whose first annual meeting this is.

Local organising committee

Scientific committee

The coordinators of the NeuralFEM project.

  • Michel Duprez — Inria, MIMESIS team, Strasbourg, France (international coordinator)
  • Franz Chouly — Universidad de la República, Montevideo, Uruguay
  • Manuela Bastidas — Universidad Nacional de Colombia, Medellín, Colombia
  • Rodolfo Araya — Universidad de Concepción, Chile
  • Virginie Ehrlacher — CERMICS, École nationale des Ponts et Chaussées, France

Contact

Questions about the workshop go to Michel Duprez, at the address in the menu.