Mathematical Sciences: Location of Outliers in Structured Data

首席研究员将研究结构化数据中异常值的检测。异常值是与大部分观测值不同的观测值。众所周知，当存在异常值时，许多参数估计程序无法产生良好的结果。鉴于此，处理异常值的一种方法是初步检测它们的存在。研究者将继续研究这个基于元素集程序的检测问题，这涉及到对数据进行许多模型的拟合，每个模型都精确拟合到原始数据的最小子集。以适当的方式结合这些元素集的信息，可以在一次传递中可靠地识别多个异常值。

The principal investigator will study the detection of outliers in structured data. Outliers are observations which appear different from the bulk of observations. It is known that many parameter estimation procedures fail to produce good results when outliers are present. In light of this, one method of handling outliers is to initially detect their presence. The investigator will continue to study this detection problem based on the elemental sets procedure, which involves fitting to the data many models, each an exact fit to a minimally-sized subset of the original data. Combining the information from these elemental sets in suitable ways, one may identify the multiple outliers reliably, and in a single pass.