DPG Phi
Verhandlungen
Verhandlungen
DPG

SKM 2021 – wissenschaftliches Programm

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

DY: Fachverband Dynamik und Statistische Physik

DY 16: Machine Learning in Dynamical Systems and Statistical Physics (joint session DY/BP)

DY 16.1: Vortrag

Freitag, 1. Oktober 2021, 11:15–11:30, H2

Tayloring Reservoir Computing Performance via Delay Time Tuning — •Tobias Hülser, Felix Köster, and Kathy Lüdge — Institut für Theoretische Physik, TU Berlin

Reservoir Computing is a versatile, fast-trainable machine learning scheme that utilises the intrinsic information-processing capacities of dynamical systems. In recent years delay-based reservoir computing emerged as a promising, easy to implement alternative to classical reservoir computing. Previous work showed that a mismatch between input time and delay time enhances computational performance significantly[1]. For delays much higher than the input time, it was shown that certain inputs cannot be recalled by the network which lead to gaps in the memory capacity[2]. Via manipulating the delays in a system of ring-coupled Stuart-Landau oscillators, we show that some of the gaps can be closed. Moreover, we can tune the range of previous inputs the reservoir can memorise. Consequently, we find a significant increase in performance for nonlinear memory tasks and the NARMA10 task.

[1] S. Stelzer et al., Neural Networks 124, 158-169 (2020)

[2] F. Köster et al., Cogn. Comput. (2020)

100% | Mobil-Ansicht | English Version | Kontakt/Impressum/Datenschutz
DPG-Physik > DPG-Verhandlungen > 2021 > SKM