{"title":"Mixed precision quantization of silicon optical neural network chip","authors":"","doi":"10.1016/j.optcom.2024.131231","DOIUrl":null,"url":null,"abstract":"<div><div>In recent years, the field of neural network research has witnessed remarkable advancements in various domains. One of the emerging approaches is the integration of photonic computing, which leverages the unique properties of light for ultra-fast information processing. In this article, we establish a mixed precision quantization model to silicon-based optical neural networks and evaluates their performance on the MNIST and Fashion-MNIST datasets. Through a genetic algorithm-based optimization process, we achieve significant parameter compression while maintaining competitive accuracy. Our findings demonstrate that with an average quantization bitwidth of 4.5 bits on the MNIST dataset, we achieve an impressive 85.94% reduction in parameter size compared to traditional 32-bit networks, with only a marginal accuracy drop of 0.65%. Similarly, on the Fashion-MNIST dataset, we achieve an average quantization bitwidth of 5.67 bits, resulting in an 82.28% reduction in parameter size with a slight accuracy drop of 0.8%.</div></div>","PeriodicalId":19586,"journal":{"name":"Optics Communications","volume":null,"pages":null},"PeriodicalIF":2.2000,"publicationDate":"2024-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Optics Communications","FirstCategoryId":"101","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0030401824009684","RegionNum":3,"RegionCategory":"物理与天体物理","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"OPTICS","Score":null,"Total":0}
引用次数: 0
Abstract
In recent years, the field of neural network research has witnessed remarkable advancements in various domains. One of the emerging approaches is the integration of photonic computing, which leverages the unique properties of light for ultra-fast information processing. In this article, we establish a mixed precision quantization model to silicon-based optical neural networks and evaluates their performance on the MNIST and Fashion-MNIST datasets. Through a genetic algorithm-based optimization process, we achieve significant parameter compression while maintaining competitive accuracy. Our findings demonstrate that with an average quantization bitwidth of 4.5 bits on the MNIST dataset, we achieve an impressive 85.94% reduction in parameter size compared to traditional 32-bit networks, with only a marginal accuracy drop of 0.65%. Similarly, on the Fashion-MNIST dataset, we achieve an average quantization bitwidth of 5.67 bits, resulting in an 82.28% reduction in parameter size with a slight accuracy drop of 0.8%.
期刊介绍:
Optics Communications invites original and timely contributions containing new results in various fields of optics and photonics. The journal considers theoretical and experimental research in areas ranging from the fundamental properties of light to technological applications. Topics covered include classical and quantum optics, optical physics and light-matter interactions, lasers, imaging, guided-wave optics and optical information processing. Manuscripts should offer clear evidence of novelty and significance. Papers concentrating on mathematical and computational issues, with limited connection to optics, are not suitable for publication in the Journal. Similarly, small technical advances, or papers concerned only with engineering applications or issues of materials science fall outside the journal scope.