Grand Séminaire OSUG

Jeudi 30 juin 2022 - 11h00

Knitted fabric, Earthquakes and Neural networks

Frédéric Lechenault - CNRS, Laboratoire de Physique Statistique, ENS Paris

— -
While studying the mechanics of a model knit, we discovered that the disorder associated with friction produces a noisy response. The statistics of this noise is similar to the Gutenberg-Richter distribution of earthquake magnitudes, and the morphology of the corresponding deformation field closely resembles seismic faults. Inspired by this analogy and our ability to gather large amounts of data in the lab, we try to forecast the events present in the response of our knit using neural networks, with the hope that our approach can be exploited in the geophysical context. In order to give a meaningful interpretation to the observed performance, we introduce a risk management framework in the form of an imaginary city subjected to the experimental noise, which is in turn solved by a reinforcement learning agent. On the one hand, this framework allows to quantitatively compare the different forecasting approaches, and on the other, it yields constructive risk management policy design.

Équipe organisatrice : Grands Séminaires ISTerre