Archiv aktualit
Special Session of the Central European Seminar In Celebration of Peter Michor’s 75th Birthday, June 6 – June 8, 2024 PDF Tisk

SCI MUNI, DMS, building 8, lecture room M1, June 6 – June 8, 2024

This mini-conference is organized in honour of Peter Michor's, who recently got his dr.h.c from our university. All the speakers are Peter's former students, they display an exciting variety of topics and they come from world top places (Kings College, Florida, ETH Zurich, etc).

All the talks will be in M1 and you are all cordially invited.


See the (preliminary) programme

Aktualizováno Úterý, 04 Červen 2024 10:29
 
Dr. Sebastien Court - Designing the monodomain model with artificial neural networks - 22.5.2004 PDF Tisk

Dne 22.5.2024 se v zasedací místnosti ÚMS od 14:00 uskuteční tato přednáška:

Dr. Sebastien Court (Institut für Mathematik, Universität Innsbruck)


Designing the monodomain model with artificial neural networks

Abstrakt:
We propose a data-driven method in order to identify the nonlinearity in the monodomain model. The monodomain model is a system coupling a  semilinear parabolic PDE with an ODE, describing the time evolution of an electric potential. Our approach provides a general answer to the problem of selecting the model when studying phenomena related to cardiac electrophysiology: From measurements, instead of determining coefficients of a prescribed nonlinearity (like the FitzHugh-Nagumo model for instance), we design the nonlinearity itself, in the form of an artificial neural network (ANN), more specifically a feedforward residual network. The relevance of this approach relies on the approximation capacities of neural networks. Training the ANN corresponds to solving an identification problem constrained by the monodomain model so parameterized. We formulate this inverse problem as an optimal control problem, and provide mathematical analysis and derivation of optimality conditions for identifying the weights of the ANN. One of the difficulties comes from the lack of smoothness of activation functions which are classically used for training deep neural networks. We will also present numerical results that demonstrate the feasibility of the strategy proposed in this work. This is joint work with Prof. Karl Kunisch (RICAM & University of Graz).

Organizuje:
doc. RNDr. Lenka Přibylová, Ph.D.
Department of Mathematics and Statistics, SCI MUNI
the Nonlinear Dynamics Team
https://science.math.muni.cz/ndteam/
Funding: European Horizon MSCA grant 101063853


Aktualizováno Pondělí, 20 Květen 2024 06:52
 
«ZačátekPředchozí12345678910DalšíKonec»