Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 49: Focus Session: Computing with Dynamical Systems: New Perspectives on Reservoirs and Applications II – Applications and Quantum RC

DY 49.1: Invited Talk

Thursday, March 21, 2024, 15:00–15:30, BH-N 243

Using reservoir computing to create surrogate models — •Lina Jaurigue — Technische Universität Ilmenau

Reservoir computing is a machine learning approach that utilizes the dynamics of a network or physical system to perform complex tasks while only training the output layer via linear regression. Due to the memory properties of the reservoir, reservoir computing is well suited to perform chaotic timeseries prediction tasks. When a reservoir is trained to perform one-step-ahead prediction of a chaotic trajectory, it can also be used to create a surrogate model for the system that the training trajectory originated from. This is done by feeding the output of the reservoir back in as input in the prediction phase. The resulting system is an autonomous dynamical system with solutions that depend on properties of the reservoir and the weights. We investigate how the properties of the reservoir, the input weights and the trained output weights influence solutions of the resulting autonomous dynamical system.

Keywords: Reservoir computing; Nonlinear dynamics; Timeseries forecasting; Bifurcation analysis; Machine learning