Recent Advances in Reservoir Computing with A Focus on Electronic Reservoirs

Abstract

Reservoir Computing is a subset of recurrent neural networks which can compute temporal-spatial tasks efficiently. In reservoir computing inputs are randomly connected to fixed untrained nodes in the reservoir layer. From the reservoir layer signals are mapped to an output layer from which they are separated into different classes. The major advantage with reservoir computing is that the training complexity is greatly simplified by training only the output layer. This way the output weights can be trained using a simple training algorithm such as a linear classifier. This allows feasibility in hardware implementations as well. The working procedure of the different subsets of reservoir computing including echo-state networks, liquid state machines and delayed feedback reservoir computing are covered in this review. This review focuses on the current trends in reservoir computing with a focus on electronic reservoir computing with analog circuits, FPGAs, VLSIs and memristors along with some applications of reservoir computing.