Mixture of experts classification using a hierarchical mixture model.pdfVIP

下载本文档

10
0
约4.81万字
约 32页
2017-04-13 发布于江苏
举报
版权申诉

Mixture of experts classification using a hierarchical mixture model.pdf

1、本文档共32页，可阅读全部内容。
2、原创力文档（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)。
5、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
6、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
7、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
8、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Mixture of experts classification using a hierarchical mixture model

Mixture of Experts Classification Using a Hierarchical Mixture Model Michalis K. Titsias and Aristidis Likas Department of Computer Science University of Ioannina 45110 Ioannina - GREECE e-mail: mtitsias@cs.uoi.gr, arly@cs.uoi.gr Abstract A three-level hierarchical mixture model for classification is pre- sented which models the following data generation process i) the data are generated by a finite number of sources (clusters) and ii) the generation mechanism of each source assumes the existence of individual internal class labelled sources (subclusters of the exter- nal cluster). The model estimates the posterior probability of class membership similar to a mixture of experts classifier. In order to learn the parameters of the model we have developed a general training approach based on maximum likelihood which results in two efficient training algorithms. Compared to other classification mixture models the proposed hierarchical model exhibits several advantages and provides improved classification performance as in- dicated by the experimental results. 1 Introduction A widely applied method for implementing the Bayes classifier is based on obtaining the posterior probabilities of class membership through the estima- tion of the class prior probabilities and the class conditional densities (Duda Hart, 1973). This is a generative approach to classification since a model of the joint distribution of the input data and the class labels is provided. The computationally intensive part of the design of such classifier concerns the es- timation of the class conditional densities. The widely used way to obtain these estimates is independently to apply density estimation methods to each class labelled data set. However, such an approach does not benefit from the existence of any common characteristics among data of different classes. For example, the data may arise from differently labelled clusters that are located in overlapping regions in the data space. A very general