简介：我们在场一个新歧管的学习算法叫了保存排列(LOPA)的本地Orthogonality。我们的算法被试图用仿射的转变排列多重本地邻居进一个全球坐标系统的本地正切空间排列(LTSA)方法启发。然而，LTSA经常没能保存象距离和角度那样的原来的几何数量。尽管为保存orthogonality的一个反复的排列过程被LTSA的作者建议，既不相应初始化也不实验被给。普罗克拉斯提斯Subspaces排列(PSA)实现由与退火模仿独立估计每旋转转变保存想法的orthogonality。然而，在PSA的优化是复杂、多重的分开的本地旋转可以生产全球性倾向於矛盾的结果。探讨这些困难，我们首先使用LTSA的伪逆诡计与统一全球坐标代表每本地直角的转变。第二，orthogonality限制被放松是半明确的编程(SDP)的一个例子。最后一个二拍子的圆舞反复的过程被采用进一步在直角的限制减少错误。广泛的实验证明LOPA能忠实地保存原来的数据集的距离，角度，内部产品，和邻居。在比较，当LOPA的运行时刻比PSA，MVU和MVE的显著地快时，LOPA的嵌入的表演比PSA的好、比得上象MVU和MVE一样的最先进的算法的。

简介：Theflowbehindathree-dimensionalrotationallyoscillatingcircularcylinderwasstudiedbyanumericalmethod.ThecomputationswereperformedataReynoldsnumberof260,whichisatalevelthattheflowwakehasdevelopedintoathree-dimensionalstatecalledMode-B.Thepurposeofthispaperistoexaminetheinfluenceofvariousrotationalamplitudes(0.1-0.7)onthewakeinstabilityoftheflow,whiletheoscillationfrequencyisfixedtothevalueofthatmeasuredinthewakeofastationarycylinder.Theresultsshowthattherotationwithsufficientlyhighamplitudebringstheflowbacktoitsnominaltwo-dimensionalstate.Moreover,itisfoundthatthevalueofthetime-averageddragandtheRMSvalueoftheliftarelargerthanthoseofastationarycircularcylinder.

简介：ExistingmanifoldlearningalgorithmsuseEuclideandistancetomeasuretheproximityofdatapoints.However,inhigh-dimensionalspace,Minkowskimetricsarenolongerstablebecausetheratioofdistanceofnearestandfarthestneighborstoagivenqueryisalmostunit.Itwilldegradetheperformanceofmanifoldlearningalgorithmswhenappliedtodimensionalityreductionofhigh-dimensionaldata.Weintroduceanewdistancefunctionnamedshrinkage-divergence-proximity(SDP)tomanifoldlearning,whichismeaningfulinanyhigh-dimensionalspace.Animprovedlocallylinearembedding(LLE)algorithmnamedSDP-LLEisproposedinlightofthetheoreticalresult.Experimentsareconductedonahyperspectraldatasetandanimagesegmentationdataset.Experimentalresultsshowthattheproposedmethodcanefficientlyreducethedimensionalitywhilegettinghigherclassificationaccuracy.

简介：

简介：

简介：