简介:Inthispaper,wegivethesensitivityanalysesbytwoapproachesforL,D,UinfactorizationA=LDUofgeneralperturbationsinAwhichsufficientlysmallinnorm.Bythematrix-vectorequationapproach,wederivethesharpcoaitionnumberforL,DandUfactors.Bythematrixequationapproachwederivecorrespondingconditionestimates.WhenAisasymmetricmatrix,thecorrespondingresultscanbeobtainedforLDLTfactorization.
简介:在这份报纸,我们学习一个乐队抑制nonnegative矩阵因式分解(乐队NMF)问题:为给定的nonnegative矩阵Y,与A作为Y斧子分解它一个nonnegative矩阵和X一个nonnegative块乐队矩阵。这个因式分解模型扩大一个单个低等级subspace模型到几重叠低等级subspaces的混合物,它能不仅提供稀少的表示,而且能从数据集捕获重要组织结构。基于重叠subspace聚类和在附近的subspaces之间的重叠的水平的俘获,二个简单、实际的算法被介绍解决乐队NMF问题。合成数据和真实图象数据的数字实验证明乐队NMF在数据表示提高NMF的性能并且处理。
简介:<正>Inthispaper,weprovethatthesetofallfactorizationindicesofacompletelypositivegraphhasnogaps.Inotherwords,wegiveanaffirmativeanswertoaquestionraisedbyN.KoganandA.Berman[8]inthecaseofcompletelypositivegraphs.
简介:Weconsiderinthispapertheproblemofrecursiveidentificationforstochasticsystemswhenthenoisemodeldoesnotsatisfythepositiverealconditionassociatedwithconvergenceofstandardalgorithms.Toavoidthepositiverealcondition,adaptivespectralfactorizationtechniquesareexploitedonthebasisofaclassofnon-standardtime-varyingrecursiveRiccatiequations.TheasymptoticpropertiesoftheRiccatiequationsarestudiedasacrucialsteptotheconvergenceresultsofthepaper.
简介:Basedonthewell-knownLeverrieralgorithm,asimpleexplicitsolutiontorightfactorizationofalinearsystemisestablished.Thissolutionisexpressedbythecontrollabilitymatrixofthegivensystemandasymmetricoperatormatrix.ApplicationsofthissolutiontoatypeofgeneralizedSylvestermatrixequationsandtheproblemofparametriceigenstructureassignmentbystatefeedbackareinvestigated,andgeneralcompleteparametricsolutionstothesetwoproblemsarededuced.Thesenewsolutionsaresimple,andpossessdesirablestructuralpropertieswhichrenderthesolutionsreadilyimplementable.Anexampledemonstratestheeffectoftheproposedresults.
简介:K1,k┐FACTORIZATIONOFBIPARTITEGRAPHSDUBEILIANGAbstract.Inthispaper,anecessaryconditionforabipartitegraphλKm,ntobeK1,k-factoriz...
简介:把复杂contourlet变换(CCT)与nonnegative矩阵因式分解(NMF)相结合的一个图象熔化方法在这份报纸被建议。在二幅图象被CCT分解以后,NMF被用于他们的highand低频率的部件分别地,一幅图象被综合。图象熔化结果的Subjective-visual-quality基于NMF和有NMF的小浪/contourlet/nonsubsampledcontourlet的联合与图象熔化方法的那些相比。试验性的结果是评估份量上,并且跑的时间也被对比。建议图象熔化方法能获得更大的信息熵,标准差和吝啬的坡度,这被显示出,它意味着它能更好从所有来源图象集成展示信息,避免背景噪音并且有效地支持在熔化图象的空间明白。
简介:声音变换(VC)基于Gaussian混合模型(GMM)是变换来源光谱指向光谱的最经典、普通的方法。然而,这个方法对敏感因为它的frame-by-frame变换在恰当上。有非否定的矩阵因式分解(NMF)的VC在这份报纸被介绍,它能阻止光谱由调整基础向量(字典)的尺寸在恰当上。以便认识到更好的非线性的印射,核NMF(KNMF)被采用完成印射的光谱。另外,增加变换的精确性,与GMM(GKNMF)相结合的KNMF也被介绍进VC。最后,KNMF,GKNMF,GMM,主要部件回归(PCR),与GMM(GPCR)相结合的PCR,部分最不方形的回归(PLSR),变弯的NMF基于关联的频率(NMF-CFW)和深神经的网络(DNN)方法与对方相比。建议GKNMF在客观评估和主观评估得到更好的性能。
简介:Thisstudywasonsuperiorityofthenon-negativematrixfactorization(NMF)algorithmforapplicationofinformationextractedwithaerialimages.First,NMFwasusedforaerialimageinformationextraction,andthenthisdatawascomparedwithaprincipalcomponentanalysis(PCA)inwhichr(thenumberofrowsorcolumnsofbasicmatrix)andEignum(thenumberofeigenvalues)weregivendifferentvalues.ExperimentalresultsshowedthattheruntimeofNMFwithr=20or50waslessthanthatofPCAwithanEignum=20or50.Also,therecognitionrateofNMFwithr=50washigherthanthatofanEignum=50.Theexperimentshowedthatnonnegativematrixfactorizationhadadvantagesofashorttimeperiodwithahighrecognitionrate.
简介:Inthispaper,wepresentanovelapproachtosynthesizingfrontalandsemi-frontalcartoon-likefacialcaricaturesfromanimage.Thecaricatureisgeneratedbywarpingtheinputfacefromtheoriginalfeaturepointstothecorrespondingexaggeratedfeaturepoints.A3Dmeanfacemodelisincorporatedtofacilitatefacetocaricaturesbyinferringthedepthof3Dfeaturepointsandthespatialtransformation.Thenthe3Dfaceisdeformedbyusingnon-negativematrixfactorizationandprojectedbacktoimageplaneforfuturewarping.Toefficientlysolvethenonlinearspatialtransformation,weproposeanovelinitializationschemetosetupLevenberg-Marquardtoptimization.Accordingtothespatialtransformation,exaggerationisappliedtothemostsalientfeaturesbyexaggeratingtheirnormalizeddifferencefromthemean.Non-photorealisticrendering(NPR)basedstylizationcompletesthecartooncaricature.Experimentsdemonstratethatourmethodoutperformsexistingmethodsintermsofviewanglesandaestheticvisualquality.
简介:Foralargeclassofdiscrete-timemultivariableplantswitharbitraryrelativedegrees,thedesignandanalysisofthedirectmodelreferenceadaptivecontrolschemeareinvestigatedunderlessrestrictiveassumptions.ThealgorithmisbasedonanewparametrizationderivedfromthehighfrequencygainmatrixfactorizationKp=LDUundertheconditionthatthesignsoftheleadingprincipalminorsofKpareknown.Byreprovingthediscrete-timeLpandL2δnormrelationshipbetweeninputsandoutputs,establishingthepropertiesofdiscrete-timeadaptivelaw,definingthenormalizingsignal,andrelatingthesignalwithallsignalsintheclosed-loopsystem,thestabilityandconvergenceofthediscrete-timemultivariablemodelreferenceadaptivecontrolschemeareanalyzedrigorouslyinasystematicfashionasinthecontinuous-timecase.