Reservoir computing uses a variety of physical, biological or computational processes to increase the non-linear diversity of input data the outputs of which (called the readout) can then be used as inputs for a simpler final machine learning stage. For example, input data might be used to drive electrical impulses into a semi-chaotic silicon substrate and the output currents measured at multiple points.
Used in glossary entries: machine learning, non-linear diversity, readout
Reservoir computing – main stages