Isotropization of Complex-Valued Reservoir Computing on 2-D Dynamical System

Reservoir computing (RC), which achieves high-speed learning through a reservoir layer with recurrent non-linear mapping to a high dimensional space, has been used for prediction and estimation of sequential data. To deal with complex amplitude data such as synthetic aperture radar (SAR) data, RC has been extended to the complex domain, complex-valued reservoir computing (CVRC). In this paper, we analyze 2-D dynamical systems and examined the isotropic dynamics of CVRC. We show that CVRC preserves the periodicity of the input signal and has a high robustness to initial values compared to non-isotropic real-valued RC (RVRC).