Powerful Encoding and Decoding Computation of Reservoir Computing

Weian Li, Huiwen Wu, Dongping Yang

Abstract: Reservoir computing (RC) has been widely applied in the fields of time series data processing and time series prediction due to its powerful data representation capability. The reservoir exhibits nonlinear dynamics, and its internal dynamics have infinitely long correlations when the system settles at the edge of chaos, rendering the system to achieve excellent computational performance. However, the encoding and decoding performance of RC is still unclear. This paper investigates the encoding and decoding abilities of the classic RC model, Echo State Network (ESN), on an image reconstruction task. The results show that ESN could greatly reconstruct grey images as well as color images, and demonstrate excellent generalization. Furthermore, a deep neural network based on ESN is proposed to resist the attacks on the trained model in experiments, showing that ESN enables the model to have excellent privacy protection ability. Our results demonstrate that the ESN’s powerful encoding and decoding computational performance makes it highly promising in facilitating tasks, such as few-shot learning and privacy computing.