- 1、本文档共51页,可阅读全部内容。
- 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
- 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载。
- 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
RecentAdvancesinClustering.ppt
Recent Advances in Clustering Barbara Hammer, Institute of Informatics, Clausthal University of Technology joint work with: Marie Cottrell (Paris), Alexander Hasenfuss (Clausthal), Alessio Micheli (Pisa), Nicolas Neubauer (Berlin), Frank-Michael Schleiff (Leipzig), Alessandro Sperduti (Padova), Mark Strickert (Gatersleben), Thomas Villmann (Leipzig) Outline Prototype based clustering Model Online/batch clustering Cost function Recent extensions Insensitivity of initialization: batch neural gas Optimum information transfer: magnification control Incorporating label information: supervision Clustering noneuclidean data: median clustering Clustering recursive data Old models Merge neural gas Prototype-based clustering … Model … Model data contained in a real-vector space prototypes characterized by locations in the data space clustering induced by the receptive fields based on the euclidean metric Online/batch clustering … Online/batch clustering Online: vector quantization init prototypes repeat present a data point adapt the winner into the direction of the data point Online/batch clustering Batch: k-means init prototypes repeat compute closest prototype for each data point set all prototypes to the centre of gravity of their points Cost function … Cost function both procedures minimize the cost function online: stochastic gradient descent ? batch: optimization w.r.t. assignments ? w.r.t. weights ? Recent extensions … Insensitivity of initialization: batch neural gas … Batch neural gas Neighborhood cooperation Batch neural gas Online neural gas: init repeat present a data point adapt all neurons according to their rank cost function Batch neural gas Batch neural gas: init repeat optimize the ranks optimize the prototypes converges in a finite number of steps, can be interpreted as Newton optimization Batch neural gas Checkerboard data: 1500 training points, 1500 test points, 2D, 100 prototypes, posterior cla
文档评论(0)