Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 48: Invited Talk Kathy Lüdge (joint session DY/SOE)
DY 48.1: Hauptvortrag
Freitag, 9. September 2022, 09:30–10:00, H19
Photonic Reservoir Computing: Analytic insights and possibilities for optimization — Lina Jaurigue1, Felix Köster2, and •Kathy Lüdge1 — 1Institute of Physics, Technische Univ. Ilmenau, Weimarer Str. 25, 98684 Ilmenau — 2Institute of Theoretical Physics, Technische Univ. Berlin, Hardenbergstr. 36, 10623 Berlin
Reservoir computing has gained a lot of attention because its relatively simple setup that can be easily implemented in hardware, specifically with optical devices. Using one nonlinear node, i.e., a laser with an optical feedback loop and time-multiplexed input, already allows to solve complex time-series prediction tasks after a proper training via linear regression. Nevertheless, the performance depends on properly adjusted timescales and not every physical system is suitable for a given task [1].
We present ways to improve the computing performance of delay-based photonic reservoir computing systems using delay-time tuning, and we discuss to what extent delay-coupled laser-networks with more than one optical element can be beneficial to improve the overall performance. Furthermore, we discuss analytic insights into the information processing capacity of a reservoir computing system and its correlation to the linear system response of the reservoir as well as to the series expansion of a chosen task.
[1] T. Hülser, F. Köster, L. C. Jaurigue, and K. Lüdge, Opt. Mater. Express 12, 1214 (2022).