Multi-functional reservoir computing

Abstract

Whereas the power of reservoir computing (RC) in inferring chaotic systems has been well established in the literature, the studies are mostly restricted to mono-functional machines where the training and testing data are acquired from the same attractor. Here, using the strategies of attractor labeling and trajectory separation, we propose a new scheme of RC capable of learning multiple attractors generated by entirely different dynamics, namely multi-functional RC. Specifically, we demonstrate that by incorporating a label channel into the standard RC, a single machine is able to learn from data the dynamics of multiple chaotic attractors, while each attractor can be accurately retrieved by inputting just a scalar in the prediction phase. The dependence of the machine performance on the labeling and separation parameters is investigated, and it is found that the machine performance is optimized when the parameters take intermediate values. The working mechanism of multi-functional RC is analyzed by the method of functional networks in neuroscience, and it is revealed that each attractor is represented by a stable, unique functional network in the reservoir, and the optimal performance arises as a balance between the stability, complexity, and distinguishability of the functional networks.