Practical Comparisons of Reservoir Topology Performance and Input Distribution in Digital Reservoir Computers

Reservoir computing has seen renewed interest as a paradigm in autonomous single and multi-agent systems due to its lightweight training and ability to utilize a wide variety of digital and physical substrates as reservoirs. Reservoir topology plays a significant role in determining the performance and dynamics of a given reservoir computer. Previous work comparing different reservoir topologies has ignored the interaction between the distribution of input nodes into the reservoir and reservoir topology. This study provides a significant contribution by comparing effects of two different input node distributions. Our results demonstrate that by concentrating input into an arbitrary region in the reservoir one is able to nearly double the linear memory performance of ring reservoir based reservoir computers.