SKM 2021 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
DY: Fachverband Dynamik und Statistische Physik
DY 16: Machine Learning in Dynamical Systems and Statistical Physics (joint session DY/BP)
DY 16.5: Talk
Friday, October 1, 2021, 12:15–12:30, H2
Investigating the role of Chaos and characteristic time scales in Reservoir Computing — Marvin Schmidt1,2, Yuriy Mokrousov1,3, Stefan Blügel1,3, Abigail Morrison2,3,4, and •Daniele Pinna1,3 — 1Peter Grünberg Institute (PGI-1), Wilhelm-Johnen-Straße, 52428 Jülich, Germany — 2Institute for Theoretical Neuroscience Institute of Neuroscience and Medicine (INM-6), Wilhelm-Johnen-Straße, 52428 Jülich, Germany — 3Institute for Advanced Simulation (IAS-6), Wilhelm-Johnen-Straße, 52428 Jülich, Germany — 4Computational and Systems Neuroscience & JARA-Institut Brain structure-function relationships (INM-10), Wilhelm-Johnen-Straße, 52428 Jülich, Germany
Reservoir Computing (RC) dynamical systems must retain information for long times and exhibit a rich representation of their driving. This talk highlights the importance of matching between input and dynamical timescales in RC systems close to chaos. We compare a chain of Fermi-Pasta-Ulam-Tsingou anharmonic oscillators and a sparsely connected network of spiking excitatory/inhibitory neurons. The first is toy model for magnetic spin-wave reservoirs while the latter that of a biological neural net. Both systems are shown to rely on a close matching of their relaxation timescales with the driving input signal's frequency in order to memorize and make precise use of the information injected. We argue that this is a general property of RC systems. We acknowledge the HGF-RSF project TOPOMANN for funding.