Tradeoffs with physical delay feedback reservoir computing

Tian Gan; Susan Stepney; Martin Albrecht Trefzer

doi:10.1109/SSCI50451.2021.9660137

Tradeoffs with physical delay feedback reservoir computing

Tian Gan, Susan Stepney, Martin Albrecht Trefzer

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

The Reservoir Computing (RC) paradigm is a supervised machine learning scheme utilizing natural computational ability of dynamical systems. Delayed chaotic systems provide rich dynamics for information processing by using the system's transient response to an external input, so have been identified as suitable systems for reservoir computing. In delay-feedback reservoir computing a virtual network is achieved through time-multiplexing, trading off input frequency and reservoir complexity. Here we investigate several features and tradeoffs that affect physical delay-dynamical reservoirs, based on the Mackey-Glass model, including the role of input mask, and a ‘devirtualisation’ of the output. We evaluate performance using the NARMA-10 benchmark. We find that random binary masks with no offset outperform other masking schemes, but require a careful choice of model parameters. We find that reading the output of a subset of ‘virtual’ nodes, the ‘devirtualised network’, does not affect computational power, which enables a tradeoff between output design complexity and readout frequency.

Original language	English
Title of host publication	2021 IEEE Symposium Series on Computational Intelligence (SSCI)
ISBN (Electronic)	978-1-7281-9048-8
DOIs	https://doi.org/10.1109/SSCI50451.2021.9660137
Publication status	Published - 5 Dec 2021
Event	2021 IEEE Symposium Series on Computational Intelligence (SSCI): International Conference on Evolvable Systems (ICES) - Florida, Orlando, United States Duration: 5 Dec 2021 → 7 Dec 2021

Conference

Conference	2021 IEEE Symposium Series on Computational Intelligence (SSCI)
Country/Territory	United States
City	Orlando
Period	5/12/21 → 7/12/21

Access to Document

10.1109/SSCI50451.2021.9660137

Cite this

@inproceedings{d66271edc6a54c54b4e829fe017025df,

title = "Tradeoffs with physical delay feedback reservoir computing",

abstract = "The Reservoir Computing (RC) paradigm is a supervised machine learning scheme utilizing natural computational ability of dynamical systems. Delayed chaotic systems provide rich dynamics for information processing by using the system's transient response to an external input, so have been identified as suitable systems for reservoir computing. In delay-feedback reservoir computing a virtual network is achieved through time-multiplexing, trading off input frequency and reservoir complexity. Here we investigate several features and tradeoffs that affect physical delay-dynamical reservoirs, based on the Mackey-Glass model, including the role of input mask, and a {\textquoteleft}devirtualisation{\textquoteright} of the output. We evaluate performance using the NARMA-10 benchmark. We find that random binary masks with no offset outperform other masking schemes, but require a careful choice of model parameters. We find that reading the output of a subset of {\textquoteleft}virtual{\textquoteright} nodes, the {\textquoteleft}devirtualised network{\textquoteright}, does not affect computational power, which enables a tradeoff between output design complexity and readout frequency.",

author = "Tian Gan and Susan Stepney and Trefzer, {Martin Albrecht}",

year = "2021",

month = dec,

day = "5",

doi = "10.1109/SSCI50451.2021.9660137",

language = "English",

isbn = "978-1-7281-9049-5",

booktitle = "2021 IEEE Symposium Series on Computational Intelligence (SSCI)",

note = "2021 IEEE Symposium Series on Computational Intelligence (SSCI) : International Conference on Evolvable Systems (ICES) ; Conference date: 05-12-2021 Through 07-12-2021",

}

TY - GEN

T1 - Tradeoffs with physical delay feedback reservoir computing

AU - Gan, Tian

AU - Stepney, Susan

AU - Trefzer, Martin Albrecht

PY - 2021/12/5

Y1 - 2021/12/5

N2 - The Reservoir Computing (RC) paradigm is a supervised machine learning scheme utilizing natural computational ability of dynamical systems. Delayed chaotic systems provide rich dynamics for information processing by using the system's transient response to an external input, so have been identified as suitable systems for reservoir computing. In delay-feedback reservoir computing a virtual network is achieved through time-multiplexing, trading off input frequency and reservoir complexity. Here we investigate several features and tradeoffs that affect physical delay-dynamical reservoirs, based on the Mackey-Glass model, including the role of input mask, and a ‘devirtualisation’ of the output. We evaluate performance using the NARMA-10 benchmark. We find that random binary masks with no offset outperform other masking schemes, but require a careful choice of model parameters. We find that reading the output of a subset of ‘virtual’ nodes, the ‘devirtualised network’, does not affect computational power, which enables a tradeoff between output design complexity and readout frequency.

AB - The Reservoir Computing (RC) paradigm is a supervised machine learning scheme utilizing natural computational ability of dynamical systems. Delayed chaotic systems provide rich dynamics for information processing by using the system's transient response to an external input, so have been identified as suitable systems for reservoir computing. In delay-feedback reservoir computing a virtual network is achieved through time-multiplexing, trading off input frequency and reservoir complexity. Here we investigate several features and tradeoffs that affect physical delay-dynamical reservoirs, based on the Mackey-Glass model, including the role of input mask, and a ‘devirtualisation’ of the output. We evaluate performance using the NARMA-10 benchmark. We find that random binary masks with no offset outperform other masking schemes, but require a careful choice of model parameters. We find that reading the output of a subset of ‘virtual’ nodes, the ‘devirtualised network’, does not affect computational power, which enables a tradeoff between output design complexity and readout frequency.

U2 - 10.1109/SSCI50451.2021.9660137

DO - 10.1109/SSCI50451.2021.9660137

M3 - Conference contribution

SN - 978-1-7281-9049-5

BT - 2021 IEEE Symposium Series on Computational Intelligence (SSCI)

T2 - 2021 IEEE Symposium Series on Computational Intelligence (SSCI)

Y2 - 5 December 2021 through 7 December 2021

ER -