Regensburg 2022 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 51: Machine Learning in Dynamics and Statistical Physics (joint session DY/SOE)

DY 51.5: Vortrag

Freitag, 9. September 2022, 11:00–11:15, H19

Investigation of plasticity in off-resonant delay-coupled reservoir computing — •Jonas Naujoks¹, Felix Köster¹, and Kathy Lüdge² — ¹Institute for Theoretical Physics, Technische Universität Berlin, 10559 Berlin, Germany — ²Institute of Physics, Technische Universität Ilmenau, Weimarer Str. 25, 98693 Ilmenau, Germany

We analyse the effect of neuronal plasticity on the performance of a delay-based reservoir computer modelled by a generic oscillator with self-feedback. The memory capacity and task-specific performance are investigated in the case of non-resonant delay-clock-cycle configurations. By modifying the temporal multiplexing of the input, the responsiveness of the virtual nodes is maximised while promoting individual decorrelation. The training is done in an unsupervised manner. The effect on the task-specific performance is investigated, while we additionally demonstrate that the memory capacity can be tuned.