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 16.4: Vortrag
Freitag, 1. Oktober 2021, 12:00–12:15, H2
Master memory function for delay-based reservoir computers — •Felix Köster1, Serhiy Yanchuk2, and Kathy Lüdge1 — 1Institut für Theoretische Physik, TU Berlin, Hardenbergstraße 36, 10623 Berlin — 2Institut für Mathematik, TU Berlin, Str. des 17. Juni 136, 10587 Berlin
The reservoir computing scheme is a versatile machine learning mechanism, which shows promising results in time-dependent task predictions in comparable fast-training times. Delay-based reservoir computing is a modification in which a single dynamical node under the influence of feedback is used as a reservoir instead of a spatially extended system.
We show that many delay-based reservoir computers considered in the literature can be characterized by a universal master memory function (MMF). Once computed for two independent parameters, this function provides linear memory capacity for any delay-based single-variable reservoir with small inputs. Moreover, we propose an analytical description of the MMF that enables its efficient and fast computation. Our approach can be applied not only to the reservoirs governed by known dynamical rules such as Mackey-Glass or Ikeda-like systems but also to reservoirs whose dynamical model is not available.