This Is Auburn

Show simple item record

Stellar Blend Image Classification Using Computationally Efficient Gaussian Processes (MuyGPs)


Metadata FieldValueLanguage
dc.contributorChinedu Anthony Eleh, cae0027@auburn.eduen_US
dc.creatorBidese, Rafael
dc.creatorEleh, Chinedu
dc.creatorZhang, Yunli
dc.creatorMolinari, Roberto
dc.creatorBillor, Nedret
dc.creatorPriest, Benjamin
dc.creatorGoumiri, Imene
dc.creatorMuyskens, Amanda
dc.creatorDunton, Alec
dc.date.accessioned2024-04-05T14:38:05Z
dc.date.available2024-04-05T14:38:05Z
dc.date.created2023-05-26
dc.identifier.urihttps://ww2.amstat.org/meetings/sdss/2023/en_US
dc.identifier.urihttps://aurora.auburn.edu/handle/11200/50639
dc.identifier.urihttp://dx.doi.org/10.35099/aurora-707
dc.description.abstractStellar blends are a challenge in visualizing celestial bodies and are typically disambiguated through expensive methods. To address this, we propose an automated pipeline to distinguish single stars and blended stars in low resolution images. We apply different normalizations to the data, which are passed as inputs into machine learning methods and to a computationally efficient Gaussian process model (MuyGPs). MuyGPs with 𝑁𝑡 ℎ root local min-max normalization achieves 86% accuracy (i.e. 12% above the second-best). Moreover, MuyGPs outperforms the benchmarked models significantly on limited training data. Further, MuyGPs low confidence predictions can be redirected to a specialist for human-assisted labeling.en_US
dc.formatPDFen_US
dc.publisherAmerican Statistical Associationen_US
dc.relation.ispartofSymposium on Data Science and Statisticsen_US
dc.rightsCC BY 4.0en_US
dc.titleStellar Blend Image Classification Using Computationally Efficient Gaussian Processes (MuyGPs)en_US
dc.typeTexten_US
dc.type.genreConference Abstracten_US
dc.description.peerreviewYesen_US
dc.locationSt. Louis, Missourien_US

Files in this item

Show simple item record