Robust Multivariate Analysis and Outlier Identification in Massive Data Sets

海量数据集中的稳健多变量分析和异常值识别

• 批准号: 9870172
• 负责人: David Rocke
• 金额: $ 10.5万
• 依托单位: University of California-Davis
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-09-01 至 2002-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9870172&HistoricalAwards=false
• 关键词: Robust; Multivariate; Analysis; Outlier; Identification

Rocke9870172This research involves new and improved methods for robust estimation of multivariate location and shape and identification of outliers. Massive data sets, especially in high dimension, form a particular target of the research. The research focuses on theoretical, methodological, and computational questions in robust multivariate analysis, and also extends the investigators previous work into cluster analysis.Identification of outliers and clusters of data in high dimension and in massive data sets is an important and difficult problem. These data sets occur in satellite imagery, in the human genome project, in x-ray astronomy, in geophysics, and in many other areas of science. They range in size from multiple gigabytes to terabytes, and present significant challenges for analysis. This project is aimed at methods for uncovering previously unknown structure in these data sets. The computationally intensive nature of the methods relates this research directly to the High Performance Computing and Communication Federal Strategic Area, and the applications to environmental science data relates it to the Environment and Global Change area. The project is supported by and associated with the National Partnership for Advanced Computational Infrastructure.

本研究涉及新的和改进的方法，用于多变量位置和形状的鲁棒估计以及离群值的识别。 海量数据，特别是高维数据，是研究的一个特殊对象。 本文的研究主要集中在稳健多元分析的理论、方法和计算问题上，同时也将研究者的工作扩展到聚类分析中，高维海量数据中异常值和聚类的识别是一个重要而又困难的问题。 这些数据集出现在卫星图像、人类基因组计划、x射线天文学、地球物理学和许多其他科学领域。 它们的大小从数千兆字节到数兆字节不等，对分析提出了重大挑战。 该项目旨在揭示这些数据集中以前未知的结构的方法。 该方法的计算密集型性质将本研究直接与高性能计算和通信联邦战略领域相关，而环境科学数据的应用将其与环境和全球变化领域相关。 该项目得到了国家高级计算基础设施伙伴关系的支持和配合。