Bayesian Training in Photonic Neural Meshes

Charis Mesaritakis; George Sarantoglou; Sergios Theodoridis; Adonis Bogris

doi:10.1109/COMPENG50184.2022.9905470

Bayesian Training in Photonic Neural Meshes

Charis Mesaritakis^*, George Sarantoglou, Sergios Theodoridis, Adonis Bogris

^*Kontaktforfatter

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

Abstract

Neural networks based on reconfigurable photonic integrated chips (RPICs) can offer zero-latency processing, marginal power consumption and operational flexibility. On the other hand, they are subject to, performance affecting, operational/fabrication deviations in their building blocks. Here, we present a Bayesian learning framework that when combined with device characterization, can dramatically decrease power consumption beyond 74% and significantly simplify the driving circuitry.

Originalsprog	Engelsk
Titel	2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022
Forlag	IEEE Signal Processing Society
Publikationsdato	2022
ISBN (Elektronisk)	9781728171241
DOI	https://doi.org/10.1109/COMPENG50184.2022.9905470
Status	Udgivet - 2022
Begivenhed	2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022 - Florence, Italien Varighed: 18 jul. 2022 → 20 jul. 2022

Konference

Konference	2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022
Land/Område	Italien
By	Florence
Periode	18/07/2022 → 20/07/2022

Navn	2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022

Bibliografisk note

Funding Information:
This work has received funding from the EU H2020 NEoteRIC project under grant agreement 871330.

Publisher Copyright:
© 2022 IEEE.

Adgang til dokumentet

10.1109/COMPENG50184.2022.9905470

AUB Link

Søg efter materialet i Aalborg Universitetsbiblioteks søgemaskine

Andre filer og links

Link to publication in Scopus

Citationsformater

@inproceedings{72e5eefd21f54040aca641a91da12e28,

title = "Bayesian Training in Photonic Neural Meshes",

abstract = "Neural networks based on reconfigurable photonic integrated chips (RPICs) can offer zero-latency processing, marginal power consumption and operational flexibility. On the other hand, they are subject to, performance affecting, operational/fabrication deviations in their building blocks. Here, we present a Bayesian learning framework that when combined with device characterization, can dramatically decrease power consumption beyond 74% and significantly simplify the driving circuitry.",

keywords = "Bayesian training, photonic neural networks, reconfigurable photonics",

author = "Charis Mesaritakis and George Sarantoglou and Sergios Theodoridis and Adonis Bogris",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022 ; Conference date: 18-07-2022 Through 20-07-2022",

year = "2022",

doi = "10.1109/COMPENG50184.2022.9905470",

language = "English",

series = "2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022",

booktitle = "2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022",

publisher = "IEEE Signal Processing Society",

address = "United States",

}

Mesaritakis, C, Sarantoglou, G, Theodoridis, S & Bogris, A 2022, Bayesian Training in Photonic Neural Meshes. i 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022. IEEE Signal Processing Society, 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022, 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022, Florence, Italien, 18/07/2022. https://doi.org/10.1109/COMPENG50184.2022.9905470

Bayesian Training in Photonic Neural Meshes. / Mesaritakis, Charis; Sarantoglou, George; Theodoridis, Sergios et al.
2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022. IEEE Signal Processing Society, 2022. (2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022).

Publikation: Bidrag til bog/antologi/rapport/konference proceeding › Konferenceartikel i proceeding › Forskning › peer review

TY - GEN

T1 - Bayesian Training in Photonic Neural Meshes

AU - Mesaritakis, Charis

AU - Sarantoglou, George

AU - Theodoridis, Sergios

AU - Bogris, Adonis

PY - 2022

Y1 - 2022

N2 - Neural networks based on reconfigurable photonic integrated chips (RPICs) can offer zero-latency processing, marginal power consumption and operational flexibility. On the other hand, they are subject to, performance affecting, operational/fabrication deviations in their building blocks. Here, we present a Bayesian learning framework that when combined with device characterization, can dramatically decrease power consumption beyond 74% and significantly simplify the driving circuitry.

AB - Neural networks based on reconfigurable photonic integrated chips (RPICs) can offer zero-latency processing, marginal power consumption and operational flexibility. On the other hand, they are subject to, performance affecting, operational/fabrication deviations in their building blocks. Here, we present a Bayesian learning framework that when combined with device characterization, can dramatically decrease power consumption beyond 74% and significantly simplify the driving circuitry.

KW - Bayesian training

KW - photonic neural networks

KW - reconfigurable photonics

UR - http://www.scopus.com/inward/record.url?scp=85141031672&partnerID=8YFLogxK

U2 - 10.1109/COMPENG50184.2022.9905470

DO - 10.1109/COMPENG50184.2022.9905470

M3 - Article in proceeding

AN - SCOPUS:85141031672

T3 - 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022

BT - 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022

PB - IEEE Signal Processing Society

T2 - 2022 IEEE Workshop on Complexity in Engineering, COMPENG 2022

Y2 - 18 July 2022 through 20 July 2022

ER -

Bayesian Training in Photonic Neural Meshes

Abstract

Konference

Bibliografisk note

Adgang til dokumentet

AUB Link

Andre filer og links

Fingeraftryk

Citationsformater