Felix Jimenez, Amanda Koepke, Mary Gregg, Michael Frey
{"title":"Generative Adversarial Network Performance in Low-Dimensional Settings.","authors":"Felix Jimenez, Amanda Koepke, Mary Gregg, Michael Frey","doi":"10.6028/jres.126.008","DOIUrl":null,"url":null,"abstract":"<p><p>A generative adversarial network (GAN) is an artifcial neural network with a distinctive training architecture, designed to create examples that faithfully reproduce a target distribution. GANs have recently had particular success in applications involving high-dimensional distributions in areas such as image processing. Little work has been reported for low dimensions, where properties of GANs may be better identifed and understood. We studied GAN performance in simulated low-dimensional settings, allowing us to transparently assess effects of target distribution complexity and training data sample size on GAN performance in a simple experiment. This experiment revealed two important forms of GAN error, tail underflling and bridge bias, where the latter is analogous to the tunneling observed in high-dimensional GANs.</p>","PeriodicalId":54766,"journal":{"name":"Journal of Research of the National Institute of Standards and Technology","volume":null,"pages":null},"PeriodicalIF":1.3000,"publicationDate":"2021-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10913853/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Research of the National Institute of Standards and Technology","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.6028/jres.126.008","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2021/1/1 0:00:00","PubModel":"eCollection","JCR":"Q3","JCRName":"INSTRUMENTS & INSTRUMENTATION","Score":null,"Total":0}
引用次数: 0
Abstract
A generative adversarial network (GAN) is an artifcial neural network with a distinctive training architecture, designed to create examples that faithfully reproduce a target distribution. GANs have recently had particular success in applications involving high-dimensional distributions in areas such as image processing. Little work has been reported for low dimensions, where properties of GANs may be better identifed and understood. We studied GAN performance in simulated low-dimensional settings, allowing us to transparently assess effects of target distribution complexity and training data sample size on GAN performance in a simple experiment. This experiment revealed two important forms of GAN error, tail underflling and bridge bias, where the latter is analogous to the tunneling observed in high-dimensional GANs.
期刊介绍:
The Journal of Research of the National Institute of Standards and Technology is the flagship publication of the National Institute of Standards and Technology. It has been published under various titles and forms since 1904, with its roots as Scientific Papers issued as the Bulletin of the Bureau of Standards.
In 1928, the Scientific Papers were combined with Technologic Papers, which reported results of investigations of material and methods of testing. This new publication was titled the Bureau of Standards Journal of Research.
The Journal of Research of NIST reports NIST research and development in metrology and related fields of physical science, engineering, applied mathematics, statistics, biotechnology, information technology.