Please use this identifier to cite or link to this item:
http://hdl.handle.net/10397/105739
DC Field | Value | Language |
---|---|---|
dc.contributor | Department of Electrical and Electronic Engineering | en_US |
dc.creator | Zhan, Y | en_US |
dc.creator | Zhang, H | en_US |
dc.creator | Lin, H | en_US |
dc.creator | Chin, LK | en_US |
dc.creator | Cai, H | en_US |
dc.creator | Karim, MF | en_US |
dc.creator | Poenar, DP | en_US |
dc.creator | Jiang, X | en_US |
dc.creator | Mak, MW | en_US |
dc.creator | Kwek, LC | en_US |
dc.creator | Liu, AQ | en_US |
dc.date.accessioned | 2024-04-15T07:45:07Z | - |
dc.date.available | 2024-04-15T07:45:07Z | - |
dc.identifier.issn | 1863-8880 | en_US |
dc.identifier.uri | http://hdl.handle.net/10397/105739 | - |
dc.language.iso | en | en_US |
dc.publisher | Wiley-VCH Verlag GmbH & Co. KGaA | en_US |
dc.subject | On-chip training | en_US |
dc.subject | Optical computing | en_US |
dc.subject | Photonic integrated chip | en_US |
dc.subject | Photonic neural networks | en_US |
dc.title | Physics-aware analytic-gradient training of photonic neural networks | en_US |
dc.type | Journal/Magazine Article | en_US |
dc.identifier.volume | 18 | en_US |
dc.identifier.issue | 4 | en_US |
dc.identifier.doi | 10.1002/lpor.202300445 | en_US |
dcterms.abstract | Photonic neural networks (PNNs) have emerged as promising alternatives to traditional electronic neural networks. However, the training of PNNs, especially the chip implementation of analytic gradient descent algorithms that are recognized as highly efficient in traditional practice, remains a major challenge because physical systems are not differentiable. Although training methods such as gradient-free and numerical gradient methods are proposed, they suffer from excessive measurements and limited scalability. State-of-the-art in situ training method is also cost-challenged, requiring expensive in-line monitors and frequent optical I/O switching. Here, a physics-aware analytic-gradient training (PAGT) method is proposed that calculates the analytic gradient in a divide-and-conquer strategy, overcoming the difficulty induced by chip non-differentiability in the training of PNNs. Multiple training cases, especially a generative adversarial network, are implemented on-chip, achieving a significant reduction in time consumption (from 31 h to 62 min) and a fourfold reduction in energy consumption, compared to the in situ method. The results provide low-cost, practical, and accelerated solutions for training hybrid photonic-digital electronic neural networks. | en_US |
dcterms.accessRights | open access | en_US |
dcterms.bibliographicCitation | Laser & photonics reviews, Apr. 2024, v. 18, no. 4, 2300445 | en_US |
dcterms.isPartOf | Laser & photonics reviews | en_US |
dcterms.issued | 2024-04 | - |
dc.identifier.scopus | 2-s2.0-85185662693 | - |
dc.identifier.eissn | 1863-8899 | en_US |
dc.identifier.artn | 2300445 | en_US |
dc.description.validate | 202404 bcwh | en_US |
dc.description.oa | Version of Record | en_US |
dc.identifier.FolderNumber | OA_TA | - |
dc.description.fundingSource | Others | en_US |
dc.description.fundingText | National Research Foundation Singapore; Ministry of Education - Singapore | en_US |
dc.description.pubStatus | Published | en_US |
dc.description.TA | Wiley (2024) | en_US |
dc.description.oaCategory | TA | en_US |
Appears in Collections: | Journal/Magazine Article |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.