Team#apmcmlz2400090

TeamNumber:

apmcmlz2400090

ProblemChosen:

A

ApplicationresearchoftheQUBOmodelinfeatureselectionandimageclassification

Summary

Withtherapiddevelopmentofquantumcomputingtechnology,itspotentialindealingwithcomplexproblemsandlarge-scaledatasetshasbecomeincreasinglyprominent,especiallyinthefieldofartificialintelligence(AI).ThisarticlefocusesonthedeepintegrationofquantumcomputingandAIandexploresitspracticalapplicationintwokeytasks:featureselectionandimageclassification.ByconvertingtheproblemintoaQUBOmodelandcombiningitwiththesimulatedannealingalgorithmprovidedbyKaiwuSDK,theperformanceoftheAImodelisoptimized,andthebottleneckproblemoftraditionalcomputingmethodsinhigh-dimensionaldataprocessinganddeeplearningmodeloptimizationissolved,providinginnovativeideasforimprovingtaskefficiencyandclassificationaccuracy.

Datapreprocessing,fortheGermancreditscoredatasetinquestion1,thispapersolvestheproblemofmissingvaluesandoutliersinthedatathroughmethodssuchasdescriptivestatistics,binning,andinterpolationfilling;nonlinearBox-Coxtransformationandfeaturenormalizationtechnologyareusedtooptimizethedatastructuretoensuretheeffectivenessoffeatureselection.Forquestion2,thispaperselectsFASHION-MNISTandCIFARdatasetsandprovideshigh-qualitydatainputforsubsequentimageclassificationmodelsthroughpreprocessingmethodssuchasimageflipping,rotation,denoising,andnormalization.

Fortask1,featureselectionoftheGermancreditscoredataset.FortheGermancreditscoredataset,thispaperconductedfeatureselectionresearchbasedontheLASSOmodelandQUBOmodel.ByconvertingthefeatureselectionproblemintoaQUBOmodel,thesimulatedannealingalgorithmofKaiwuSDKisusedforoptimizationandsolution.TheresultsshowthattheQUBOmodelissignificantlybetterthan