LGF Seminar – Renaud Ferrier – September 21, 2026

An LGF seminar will take place on Monday, September 21, 2026 at 2:00 PM in room J3.02, presented by Renaud Ferrier, researcher-lecturer in the MPE team.

Title

Low-rank estimates of the sensitivity matrix: how to identify the parameters of our models without adjoint problem or perturbative calculation?

Renaud FERRIER, Paul BARAL

Abstract

In the context of parametric identification, we focus on a method consisting of approximating sensitivities (partial derivatives of observables) using a low-rank operator. In particular, we use the Broyden update method. This approach theoretically promises a very low CPU cost per iteration (only one direct calculation per iteration) for an extremely reduced implementation cost (compared in particular to setting up an adjoint problem). In this contribution, we propose to evaluate the approach on problems of increasing complexity and realism.

Keywords: Broyden update, Finite Elements Models Updating (FEMU), parameter identification, computation time reduction, inverse problems, Gauss-Newton algorithm