Litcius/Paper detail

Enhancing computational performance of a spin-wave reservoir computer with input synchronization

Stuart Watt, Mikhail Kostylev, А. Б. Устинов

2021Journal of Applied Physics46 citationsDOI

Abstract

A spin-wave delay-line active-ring oscillator has recently been proposed as a suitable substrate to implement the physical reservoir computing model. The concept displays the required properties of fading memory and nonlinearity characteristic to the model. In this paper, we improve the concept by increasing the signal delay time in the yttrium-iron garnet film by more than four times, and we examine further the improved system by evaluating experimentally the performance on two benchmark classification tasks. The short-term memory (STM) task evaluates the linear memory characteristics of the RC, while the parity-check (PC) task evaluates the nonlinear computing capability. Adequate performance on both is achieved, and the linear memory is shown to be strongly dependent on the synchronization between the reservoir computer (RC) inputs and the active-ring circulation time. The extended delay time and other major improvements result in STM and PC capacities reaching maximum values of 4.68 and 1.74, respectively.

Topics & Concepts

Reservoir computingSynchronization (alternating current)Computer scienceBenchmark (surveying)Nonlinear systemTask (project management)Yttrium iron garnetElectronic engineeringPhysicsEngineeringTelecommunicationsOpticsArtificial neural networkQuantum mechanicsChannel (broadcasting)Recurrent neural networkMachine learningGeographySystems engineeringGeodesyNeural Networks and Reservoir ComputingAdvanced Memory and Neural ComputingOptical Network Technologies
Enhancing computational performance of a spin-wave reservoir computer with input synchronization | Litcius