Rausell-Campo, JR.; Pérez-López, D. (2022). Reconfigurable Activation Functions in Integrated Optical Neural Networks. IEEE Journal of Selected Topics in Quantum Electronics. 28(4):1-13. https://doi.org/10.1109/JSTQE.2022.3169833
Por favor, use este identificador para citar o enlazar este ítem: http://hdl.handle.net/10251/194394
Title:
|
Reconfigurable Activation Functions in Integrated Optical Neural Networks
|
Author:
|
Rausell-Campo, Jose Roberto
Pérez-López, Daniel
|
UPV Unit:
|
Universitat Politècnica de València. Instituto Universitario de Telecomunicación y Aplicaciones Multimedia - Institut Universitari de Telecomunicacions i Aplicacions Multimèdia
|
Issued date:
|
|
Abstract:
|
[EN] The implementation of nonlinear activation functions is one of the key challenges that optical neural networks face. To the date, different approaches have been proposed, including switching to digital implementations, ...[+]
[EN] The implementation of nonlinear activation functions is one of the key challenges that optical neural networks face. To the date, different approaches have been proposed, including switching to digital implementations, electro-optical or all optical. In this article, we compare the response of different electro-optic architectures where part of the input optical signal is converted into the electrical domain and used to self-phase modulate the intensity of the remaining optical signal. These architectures are made up of Mach Zehnder Interferometers (MZI) and microring resonators (MRR). We have compared the corresponding transfer functions with commonly used activation functions in state-of-the-art machine learning models and carried out an in-depth analysis of the capabilities of those architectures to generate the proposed activation functions. We demonstrate that a ring assisted MZI and a two-ring assisted MZI present the highest expressivity among the proposed structures. To the best of our knowledge, this is the first time that a quantified analysis of the capabilities of optical devices to mimic state-of-the-art activation functions is presented. The obtained activation functions are benchmarked on two machine learning examples: classification task using the Iris dataset, and image recognition using the MNIST dataset. We use complex-valued feed-forward neural networks and get test accuracies of 97% and 95% respectively.
[-]
|
Subjects:
|
Optical interferometry
,
Nonlinear optics
,
Optical imaging
,
Optical modulation
,
Optical signal processing
,
Optical bistability
,
Adaptive optics
,
Complex-valued neural networks
,
Electro-optic modulation
,
Machine learning
,
Optical activation functions
,
Optical neural networks
|
Copyrigths:
|
Reconocimiento (by)
|
Source:
|
IEEE Journal of Selected Topics in Quantum Electronics. (issn:
1077-260X
)
|
DOI:
|
10.1109/JSTQE.2022.3169833
|
Publisher:
|
Institute of Electrical and Electronics Engineers
|
Publisher version:
|
https://doi.org/10.1109/JSTQE.2022.3169833
|
Coste APC:
|
3025 €
|
Project ID:
|
info:eu-repo/grantAgreement/EC/H2020/871330/EU
info:eu-repo/grantAgreement/UPV-VIN//PAID-01-20-24//Fotónica integrada programable para aplicaciones de inteligencia artificial./
|
Thanks:
|
This work was supported in part by FPI-UPV Grant Program under Grant PAID-01-20-24 from the Universitat Politecnica de Valencia, through the Spanish MINECO Juan de la Cierva Program and in part by the H2020-ICT2019-2 ...[+]
This work was supported in part by FPI-UPV Grant Program under Grant PAID-01-20-24 from the Universitat Politecnica de Valencia, through the Spanish MINECO Juan de la Cierva Program and in part by the H2020-ICT2019-2 Neoteric 871330 Project.
[-]
|
Type:
|
Artículo
|