资源描述:
《[分析] convex optimization — boyd & vandenberghe》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ConvexOptimization—Boyd&Vandenberghe1.Introduction•mathematicaloptimization•least-squaresandlinearprogramming•convexoptimization•example•coursegoalsandtopics•nonlinearoptimization•briefhistoryofconvexoptimization1–1Mathematicaloptimization(mathematical)optimizati
2、onproblemminimizef0(x)subjecttofi(x)≤bi,i=1,...,m•x=(x1,...,xn):optimizationvariablesn•f0:R→R:objectivefunctionn•fi:R→R,i=1,...,m:constraintfunctionsoptimalsolutionx⋆hassmallestvalueoffamongallvectorsthat0satisfytheconstraintsIntroduction1–2Examplesportfoliooptim
3、ization•variables:amountsinvestedindifferentassets•constraints:budget,max./min.investmentperasset,minimumreturn•objective:overallriskorreturnvariancedevicesizinginelectroniccircuits•variables:devicewidthsandlengths•constraints:manufacturinglimits,timingrequirement
4、s,maximumarea•objective:powerconsumptiondatafitting•variables:modelparameters•constraints:priorinformation,parameterlimits•objective:measureofmisfitorpredictionerrorIntroduction1–3Solvingoptimizationproblemsgeneraloptimizationproblem•verydifficulttosolve•methodsinvol
5、vesomecompromise,e.g.,verylongcomputationtime,ornotalwaysfindingthesolutionexceptions:certainproblemclassescanbesolvedefficientlyandreliably•least-squaresproblems•linearprogrammingproblems•convexoptimizationproblemsIntroduction1–4Least-squaresminimizekAx−bk22solving
6、least-squaresproblems•analyticalsolution:x⋆=(ATA)−1ATb•reliableandefficientalgorithmsandsoftware•computationtimeproportionalton2k(A∈Rk×n);lessifstructured•amaturetechnologyusingleast-squares•least-squaresproblemsareeasytorecognize•afewstandardtechniquesincreaseflexi
7、bility(e.g.,includingweights,addingregularizationterms)Introduction1–5LinearprogrammingminimizecTxsubjecttoaTx≤b,i=1,...,miisolvinglinearprograms•noanalyticalformulaforsolution•reliableandefficientalgorithmsandsoftware•computationtimeproportionalton2mifm≥n;lesswith
8、structure•amaturetechnologyusinglinearprogramming•notaseasytorecognizeasleast-squaresproblems•afewstandardtricksusedtoconvertproblemsintolinearprograms(e.g.,pr