apmcmlz2400090亚太杯分赛道五岳杯一等奖.pdf

Team#apmcmlz2400090

TeamNumber:apmcmlz2400090

ProblemChosen:A

ApplicationresearchoftheQUBOmodelinfeatureselectionand

imageclassification

Summary

Withtherapiddevelopmentofquantumcomputingtechnology,itspotentialindealingwith

complexproblemsandlarge-scaledatasetshasbecomeincreasinglyprominent,especiallyinthe

fieldofartificialintelligence(AI).Thisarticlefocusesonthedeepintegrationofquantumcomputing

andAIandexploresitspracticalapplicationintwokeytasks:featureselectionandimageclassifi-

cation.ByconvertingtheproblemintoaQUBOmodelandcombiningitwiththesimulatedanneal-

ingalgorithmprovidedbyKaiwuSDK,theperformanceoftheAImodelisoptimized,andthebot-

tleneckproblemoftraditionalcomputingmethodsinhigh-dimensionaldataprocessinganddeep

learningmodeloptimizationissolved,providinginnovativeideasforimprovingtaskefficiencyand

classificationaccuracy.

Datapreprocessing,fortheGermancreditscoredatasetinquestion1,thispapersolvesthe

problemofmissingvaluesandoutliersinthedatathroughmethodssuchasdescriptivestatistics,

binning,andinterpolationfilling;nonlinearBox-Coxtransformationandfeaturenormalizationtech-

nologyareusedtooptimizethedatastructuretoensuretheeffectivenessoffeatureselection.For

question2,thispaperselectsFASHION-MNISTandCIFARdatasetsandprovideshigh-qualitydata

inputforsubsequentimageclassificationmodelsthroughpreprocessingmethodssuchasimageflip-

ping,rotation,denoising,andnormalization.

Fortask1,featureselectionoftheGermancreditscoredataset.FortheGermancreditscore

dataset,thispaperconductedfeatureselectionresearchbasedontheLASSOmodeland