Asymmetrically connected reservoir networks learn better

TL;DR

This paper demonstrates that reservoir networks with asymmetric and random connectivity outperform structured and biologically inspired topologies in time series prediction tasks, highlighting the importance of connectivity asymmetry for computational power.

Contribution

It systematically investigates how different connectivity structures affect reservoir network performance, revealing that asymmetry and randomness enhance computational capacity.

Findings

Asymmetric and random reservoirs outperform structured ones in time series prediction.

Information processing capacity is highest in asymmetric, randomly connected reservoirs.

Structured topologies like small-world are less effective for the tested task.

Abstract

We show that connectivity within the high-dimensional recurrent layer of a reservoir network is crucial for its performance. To this end, we systematically investigate the impact of network connectivity on its performance, i.e., we examine the symmetry and structure of the reservoir in relation to its computational power. Reservoirs with random and asymmetric connections are found to perform better for an exemplary Mackey-Glass time series than all structured reservoirs, including biologically inspired connectivities, such as small-world topologies. This result is quantified by the information processing capacity of the different network topologies which becomes highest for asymmetric and randomly connected networks.