Reservoir computing based on quenched chaos

Abstract

Reservoir computing (RC) is a brain-inspired computing framework that employs a transient dynamical system whose reaction to an input signal is transformed to a target output. One of the central problems in RC is to find a reliable reservoir with a large criticality, since computing performance of a reservoir is maximized near the phase transition. In this work, we propose a continuous reservoir that utilizes transient dynamics of coupled chaotic oscillators in a critical regime where sudden amplitude death occurs. This "explosive death" not only brings the system a large criticality which provides a variety of orbits for computing, but also stabilizes them which otherwise diverge soon in chaotic units. The proposed framework shows better results in tasks for signal reconstructions than RC based on explosive synchronization of regular phase oscillators. We also show that the information capacity of the reservoirs can be used as a predictive measure for computational capability of a reservoir at a critical point.

Existence of reservoir with finite-dimensional output for universal reservoir computing

“…This result suggests that chaos is key to the reservoir’s universality. In practice, chaotic reservoirs have already been considered in research and are known to be useful 21 , 22 .…”

Semiconductor technologies and related topics for implementation of electronic reservoir computing systems

“…The biological system is considered to use, in part, the dynamic nodes dealing with spontaneous activity [62][63][64]. From the viewpoint of generating complex dynamics, a chaotic oscillator is attractive and has been investigated for the reservoir [65,66]. To satisfy the requisite of consistency, it is necessary to implement a mechanism to control the phases of all nodes.…”

“…The 'edge of chaos' has been discussed recently in terms of computing capacity [65,66,87]. A nonlinear system generating chaos has a boundary between the deterministic and chaotic behaviors that depends on the system parameters.…”

“…The two behaviors can be switched by the external signal. Furthermore, the bounded instability appears in the transition region [65,66,88] and it potentially provides another class of complex dynamics that contributes to the increased computing performance of RC.…”

Synchronization of non-smooth chaotic systems via an improved reservoir computing

“…Among machine learning methods, reservoir computing (RC) method 19 , 20 also called Echo State Network (ESN) and Liquid State Machine (LSM) by Jaeger 21 and Maass 22 respectively, greatly attracts the attention of researchers due to its superior learning and prediction ability, and has been efficiently applied to the prediction of phase coherence 23 , speech recognition 24 , extreme event 25 , and other dynamical behaviors. Furtherly, some improved methods 26 – 30 of RC are also proposed for enhancing the ability in forecasting. For instance, Haluszczynski et al 27 improve the prediction performance of RC via reducing network size of the reservoir.…”