Regensburg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 45: Poster Session: Nonlinear Dynamics, Pattern Formation, Data Analytics and Machine Learning
DY 45.11: Poster
Donnerstag, 8. September 2022, 15:00–18:00, P2
Reservoir computing with memory cells: Impact of perturbations and phase effects — •Noah Jaitner1, Elizabeth Robertson3, Lina Jaurigue2, Janik Wolters3, and Kathy Lüdge2 — 1Institute of Theoretical Physics, Technische UniversitätBerlin, Hardenbergstr. 36, 10623 Berlin, Germany — 2Institut f. Physik, Technische Universität Ilmenau, Weimarer Str. 25, 98684 Ilmenau, Germany — 3Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Insitut für Optische Sensorsysteme, Rutherfordstr. 2, 12489 Berlin, Germany
Reservoir computing is a versatile, fast-trainable approach for machine learning that utilities the capabilities of dynamical systems. The common approach is to use a nonlinear element and a delay line to construct a virtual network. These virtual networks have limited topology. By utilizing cesium cells as coherent optical memory cells to create a hybrid architecture [1] the limitations of topology can be overcome and a more dynamically versatile virtual network can be created. The optical memory used in the corresponding experiment performs well in memory bandwidth but experiences high noise levels. [2] The dynamics and noise resistance of this approach is examined to find an optimal approach for different time series prediction tasks.
[1] L. C. Jaurigue, E. Robertson, J. Wolters and K. Lüdge Entropy 23, 1099-4300 (2021).
[2] L. Esguerra, L. Meßner, E. Robertson, N. V. Ewald, M. Gündoan and J. Wolters arXiv:2203.06151.