Multifunctionality in a Reservoir Computer

TL;DR

This paper demonstrates how reservoir computing can be trained to perform multiple tasks simultaneously, mimicking biological neural network multifunctionality, and analyzes the effects of parameter changes and untrained attractors.

Contribution

It introduces a training method for reservoir computers to achieve multifunctionality and analyzes untrained attractors and bifurcations affecting performance.

Findings

Reservoir computers can perform multiple tasks without changing connections.

Parameter variations significantly impact multifunctionality.

Existence of untrained attractors within the prediction space.

Abstract

Multifunctionality is a well observed phenomenological feature of biological neural networks and considered to be of fundamental importance to the survival of certain species over time. These multifunctional neural networks are capable of performing more than one task without changing any network connections. In this paper we investigate how this neurological idiosyncrasy can be achieved in an artificial setting with a modern machine learning paradigm known as `Reservoir Computing'. A training technique is designed to enable a Reservoir Computer to perform tasks of a multifunctional nature. We explore the critical effects that changes in certain parameters can have on the Reservoir Computers' ability to express multifunctionality. We also expose the existence of several `untrained attractors'; attractors which dwell within the prediction state space of the Reservoir Computer that were not part of the training. We conduct a bifurcation analysis of these untrained attractors and discuss the implications of our results.