Phase Transition Adaptation

TL;DR

This paper introduces Phase Transition Adaptation, a local unsupervised learning method for Reservoir Computing that steers system dynamics to the edge of stability, enhancing computational capacity across various datasets.

Contribution

It presents a novel unsupervised learning mechanism that improves reservoir dynamics by targeting the edge of stability, extending the original Reservoir Computing approach.

Findings

Consistently improves computational capacity

Effective across multiple datasets

Steers system dynamics to the edge of stability

Abstract

Artificial Recurrent Neural Networks are a powerful information processing abstraction, and Reservoir Computing provides an efficient strategy to build robust implementations by projecting external inputs into high dimensional dynamical system trajectories. In this paper, we propose an extension of the original approach, a local unsupervised learning mechanism we call Phase Transition Adaptation, designed to drive the system dynamics towards the `edge of stability'. Here, the complex behavior exhibited by the system elicits an enhancement in its overall computational capacity. We show experimentally that our approach consistently achieves its purpose over several datasets.