亚太杯数学建模优秀论文apmcmlz2400105.pdfVIP

Team#apmcmlz2400105

TeamNumber:apmcmlz2400105

ProblemChosen:

2024APMCMWuyueCupsummarysheet

Optimizationproblemsinthefeatureselectoranddeeplearningleadtohighermetrics

inmostcases.Howeverclassicalalgorithmsbasedonheuristicsareusuallycomplexto

findtheoptimalsolution.Fortunately,coherentopticalquantumcomputersbasedon

QUBOsolvingareabletoquicklysearchforoptimalsolutionsinexponentialHilbert

space.Inthispaper,wesolveindividualproblemsbyconstructingQUBO-basedmodels

andsolvingtheminsimulatedannealingbykaiwuSDKandCPQC-1quantumcomputer

byQBoson.

Problem1.

A)InovativeMethod:TheworkflowdrawinFig.(1).First,weestablishacomprehen-

sivefeatureevaluationusingKalmanfiltering,whichisabletoestimateinter-feature

interactionsandfeature-labelcorrelationswithhighaccuracy.Second,thisinter-

actionmatrixisusedforclusteringanddimensionalityreductiontotransformthe

featurecombinationproblemintoaclustercombinationproblem,whicheffec-

tivelyreducesthespatiotemporalcomplexityoftheQUBOmodel.Finally,the

QUBOmodelisbuiltbasedonmaximizingfeatureindependencewhilereducing

thenumberofselectedfeatures.

B)BestResult:UsingourownCQFSmethod,theoptimalAUCfortheclassification

taskcanreach0.91,and0.81evenwhenthequbitsis3.

C)FutureValues:Asquantumcomputinghardwareandalgorithmsadvance,itis

anticipatedthattheCQFSmethodwillofferamoreefficientsolutionforhigh-

dimensionalfeatureselectionandlarge-scaledataanalysis,therebyfacilitatingthe

explorationandapplicationofmoreoptimalclassificationmodels.

Problem2.

A)InovativeMethod:TheworkflowdrawinFig.(3).First,aclassical-quantumhy