Fig. 5: Time-series information processing task implementations.
From: Generative complex networks within a dynamic memristor with intrinsic variability
MCSTM (a) and MCPC (b) of a single PBAONC complex network reservoir implemented with a dynamic memristor. MCSTM (c) and MCPC (d) of a simple multi-device reservoir set (time-slot assignments are the same for each component reservoir) containing 600 virtual nodes in total (each component reservoir contains N virtual nodes). Comparison of MCSTM (e) and MCPC (f) among four mixed reservoir sets, each containing 600 virtual nodes in total but a different number of component reservoirs (therefore different N). Within a set, the time-slot assignments (θs) for each component reservoir are not all the same (therefore different Dmax). g The 64 frequency-channel cochleogram of spoken digit “nine” taken from a female speaker, preprocessed by the Lyon passive ear model (left). h Confusion matrix showing the experimentally obtained classification results from the memristive mixed PBAONC complex network reservoir versus the correct outputs.
