Efficient on-chip training of optical neural networks using genetic algorithm
Recent advances in silicon photonic chips have made huge progress in optical computing owing to their flexibility in the reconfiguration of various tasks. Its deployment of neural networks serves as an alternative for mitigating the rapidly increased demand for computing resources in electronic plat...
Saved in:
Main Authors: | , , , , , , , , , , , , , |
---|---|
Other Authors: | |
Format: | Article |
Language: | English |
Published: |
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/151456 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
id |
sg-ntu-dr.10356-151456 |
---|---|
record_format |
dspace |
spelling |
sg-ntu-dr.10356-1514562021-06-21T05:41:44Z Efficient on-chip training of optical neural networks using genetic algorithm Zhang, Hui Thompson, Jayne Gu, Mile Jiang, Xudong Cai, Hong Liu, Patricia Yang Shi, Yuzhi Zhang, Yi Muhammad Faeyz Karim Lo, Guo Qiang Luo, Xianshu Dong, Bin Kwek, Leong Chuan Liu, Ai Qun School of Electrical and Electronic Engineering School of Physical and Mathematical Sciences Engineering Optical Neural Networks On-chip Training Recent advances in silicon photonic chips have made huge progress in optical computing owing to their flexibility in the reconfiguration of various tasks. Its deployment of neural networks serves as an alternative for mitigating the rapidly increased demand for computing resources in electronic platforms. However, it remains a formidable challenge to train the online programmable optical neural networks efficiently, being restricted by the difficulty in obtaining gradient information on a physical device when executing a gradient descent algorithm. Here, we experimentally demonstrate an efficient, physics-agnostic, and closed-loop protocol for training optical neural networks on chip. A gradient-free algorithm, that is, the genetic algorithm, is adopted. The protocol is on-chip implementable, physical agnostic (no need to rely on characterization and offline modeling), and gradient-free. The protocol works for various types of chip structures and is especially helpful to those that cannot be analytically decomposed and characterized. We confirm its viability using several practical tasks, including the crossbar switch and the Iris classification. Finally, by comparing our physics-agonistic and gradient-free method to the off-chip and gradient-based training methods, we demonstrate the robustness of our system to perturbations such as imperfect phase implementation and photodetection noise. Optical processors with gradient-free genetic algorithms have broad application potentials in pattern recognition, reinforcement learning, quantum computing, and realistic applications (such as facial recognition, natural language processing, and autonomous vehicles). Ministry of Education (MOE) National Research Foundation (NRF) Accepted version This work was supported by the Singapore Ministry of Education (MOE) Tier 3 grant (MOE2017-T3-1-001), the Singapore National Research Foundation (NRF) National Natural Science Foundation of China (NSFC) joint grant (NRF2017NRF-NSFC002-014), the Singapore National Research Foundation under the Competitive Research Program (NRF-CRP13-2014-01), the NRF Fellowship reference no. NRF-NRFF2016-02. 2021-06-21T05:41:44Z 2021-06-21T05:41:44Z 2021 Journal Article Zhang, H., Thompson, J., Gu, M., Jiang, X., Cai, H., Liu, P. Y., Shi, Y., Zhang, Y., Muhammad Faeyz Karim, Lo, G. Q., Luo, X., Dong, B., Kwek, L. C. & Liu, A. Q. (2021). Efficient on-chip training of optical neural networks using genetic algorithm. ACS Photonics, 8(6), 1662-1672. https://dx.doi.org/10.1021/acsphotonics.1c00035 2330-4022 https://hdl.handle.net/10356/151456 10.1021/acsphotonics.1c00035 6 8 1662 1672 en ACS Photonics This document is the Accepted Manuscript version of a Published Work that appeared in final form in ACS Photonics, copyright © American Chemical Society after peer review and technical editing by the publisher. To access the final edited and published work see https://doi.org/10.1021/acsphotonics.1c00035 application/pdf |
institution |
Nanyang Technological University |
building |
NTU Library |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
NTU Library |
collection |
DR-NTU |
language |
English |
topic |
Engineering Optical Neural Networks On-chip Training |
spellingShingle |
Engineering Optical Neural Networks On-chip Training Zhang, Hui Thompson, Jayne Gu, Mile Jiang, Xudong Cai, Hong Liu, Patricia Yang Shi, Yuzhi Zhang, Yi Muhammad Faeyz Karim Lo, Guo Qiang Luo, Xianshu Dong, Bin Kwek, Leong Chuan Liu, Ai Qun Efficient on-chip training of optical neural networks using genetic algorithm |
description |
Recent advances in silicon photonic chips have made huge progress in optical computing owing to their flexibility in the reconfiguration of various tasks. Its deployment of neural networks serves as an alternative for mitigating the rapidly increased demand for computing resources in electronic platforms. However, it remains a formidable challenge to train the online programmable optical neural networks efficiently, being restricted by the difficulty in obtaining gradient information on a physical device when executing a gradient descent algorithm. Here, we experimentally demonstrate an efficient, physics-agnostic, and closed-loop protocol for training optical neural networks on chip. A gradient-free algorithm, that is, the genetic algorithm, is adopted. The protocol is on-chip implementable, physical agnostic (no need to rely on characterization and offline modeling), and gradient-free. The protocol works for various types of chip structures and is especially helpful to those that cannot be analytically decomposed and characterized. We confirm its viability using several practical tasks, including the crossbar switch and the Iris classification. Finally, by comparing our physics-agonistic and gradient-free method to the off-chip and gradient-based training methods, we demonstrate the robustness of our system to perturbations such as imperfect phase implementation and photodetection noise. Optical processors with gradient-free genetic algorithms have broad application potentials in pattern recognition, reinforcement learning, quantum computing, and realistic applications (such as facial recognition, natural language processing, and autonomous vehicles). |
author2 |
School of Electrical and Electronic Engineering |
author_facet |
School of Electrical and Electronic Engineering Zhang, Hui Thompson, Jayne Gu, Mile Jiang, Xudong Cai, Hong Liu, Patricia Yang Shi, Yuzhi Zhang, Yi Muhammad Faeyz Karim Lo, Guo Qiang Luo, Xianshu Dong, Bin Kwek, Leong Chuan Liu, Ai Qun |
format |
Article |
author |
Zhang, Hui Thompson, Jayne Gu, Mile Jiang, Xudong Cai, Hong Liu, Patricia Yang Shi, Yuzhi Zhang, Yi Muhammad Faeyz Karim Lo, Guo Qiang Luo, Xianshu Dong, Bin Kwek, Leong Chuan Liu, Ai Qun |
author_sort |
Zhang, Hui |
title |
Efficient on-chip training of optical neural networks using genetic algorithm |
title_short |
Efficient on-chip training of optical neural networks using genetic algorithm |
title_full |
Efficient on-chip training of optical neural networks using genetic algorithm |
title_fullStr |
Efficient on-chip training of optical neural networks using genetic algorithm |
title_full_unstemmed |
Efficient on-chip training of optical neural networks using genetic algorithm |
title_sort |
efficient on-chip training of optical neural networks using genetic algorithm |
publishDate |
2021 |
url |
https://hdl.handle.net/10356/151456 |
_version_ |
1703971230727536640 |