Evolving carbon nanotube reservoir computers

Abstract

Reservoir Computing is a useful general theoretical model for many dynamical systems. Here we show the first steps to applying the reservoir model as a simple computational layer to extract exploitable information from physical substrates consisting of single-walled carbon nanotubes and polymer mixtures. We argue that many physical substrates can be represented and configured into working reservoirs given some pre-training through evolutionary selected input-output mappings and targeted input stimuli.