Reservoir Computing in Embedded Systems: Three variants of the reservoir algorithm
Nicholas Soures, Corey Merkel, Dhireesha Kudithipudi, Clare Thiem, Nathan McDonald · IEEE Consumer Electronics Magazine · 2017
The rich computational dynamics coupled with simple training in reservoir computing (RC) algorithms makes them a natural choice for spatiotemporal learning systems in embedded platforms. In this article, we study three variants of the reservoir algorithm: the echo-state network (ESN), the liquid-state machine (LSM), and the time-delay reservoirs (TDRs). A digital reservoir architecture is used as a baseline test bed to benchmark for epileptic seizure detection. An average accuracy of 85% is observed across the three variants of the algorithm and the effectiveness of the reservoir is assessed with different metrics.