模式分类学习课件pr10part2_ding.docxVIP

 Patter n Classification All materials in these slides were taken from Pattern Classification (2nd ed) by R? O. Dudaf P? E Hart and D. G. Stork, John Wiley Sons, 2000 with the permission of the authors and the publisher Data Description Clustering ■ Learning the Structures of multidimensional patter ns from a set of un labeled samples 分 If we know that data come from a specific distribution, such data can be represented by a compact set of parameters {sufficient statistic^) ? If samples are considered coming from a specific distribution, but actually they are not, these statistics is a misleading representation of the data PAGE PAGE # If little prior knowledge can be assumed, the assumption of a parametric form is meaningless: we are actually imposing structure on data, not finding structure on ! it!In these cases, one can use non parametric methods to estimate the unknown mixture density If the goal is to find subclasses, one can use a clustering procedure to identify groups of data points having strong internal similarities ■ Clustering: clustering procedures yield a data description in terms of clusters or groups of data points that possess strong internal similarities ; - I Similarity measures B The question is how to evaluate that the samples in one cluster are more similar-among them than，/ : ^samplesJnother clusters. - TOC \o 1-5 \h \z ，“ J X : . ? :z 連卵crisses： i :一，， ‘ , jfeHow to measure the similarity between samples? Bn How to evaluate a partitioning of a set into clusters? 厂.. \JY\e most obvious [neasure of similarity (or: 4 么ssi叩ilarity) between samples js^e distance between宅陶也工乡血欣月metric.「:厂?… ； h Once defined^this measu^, x)he would； expect : 空 distance between samplfesthe same clustet t6 be significantly less than the distanee between samples in different classes? Euclidean distance is a possible metric: a possible criterion is to assume samples belonging to same cluster if their distanee is less than a threshold d° Xj Xj FIGURE IH.7?