Dresden 2026 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 6: Machine Learning in Dynamics and Statistical Physics I
DY 6.5: Vortrag
Montag, 9. März 2026, 10:30–10:45, HÜL/S186
Understanding task performance of time-multiplexed optical reservoir computing via polynomial expansion — •Elias Koch1, Julien Javaloyes2, Svetlana V. Gurevich1,3, and Lina Jaurigue4 — 1Institute for Theoretical Physics, University of Münster, Wilhelm-Klemm-Str.9 48149 Münster, Germany — 2Departament de Física and IAC3, Universitat de les Illes Balears, Campus UIB 07122 Mallorca, Spain — 3Center for Data Science and Complexity (CDSC), University of Münster, Corrensstrasse 2, Münster, 48149, Germany — 4Institute of Physics, Technische Universität Ilmenau , 98693 Ilmenau, Germany
We study the dynamics of a reservoir computer, realized as a linear optical microcavity with a time-multiplexed injection stream. In the first step, the output is processed with different nonlinearities, allowing to analyze the resulting polynomials and to what extend they can approximate different tasks. To that end, we compare two different discrete tasks, both derived from the Lorenz system through integration with a Runge-Kutta (4) scheme, but sampled to different stepsizes. There, we identify the respective underlying polynomial map and discuss the occuring terms. We compare these results with the impact of employing nonlinear nodes by introducing a Kerr nonlinearity in the optical microcavity.
Keywords: Reservoir Computing; nonlinear; optics; laser; machine learning