Perspectives on High-Dimensional Data Analysis

• Activity Number: 556
• Type: Roundtables
• Date/Time: Wednesday, August 7, 2013 : 12:30 PM to 1:50 PM
• Sponsor: Quality and Productivity Section
• Author(s): Ejaz Syed Ahmed*+
• Companies: Brock University
• Keywords: High Dimesional Data ; Absolute Penalty Estimation ; Shrinakage Estimation ; Variable Selection ; Regularization ; Estimation consistency

Abstract

We will dicuss the various estimation strategies based on absolute penalty estimation (APE).The continued rapid advancement of modern technology is allowing scientists to collect data of increasingly unprecedented size and complexity. Examples include epigenomic data, genomic data, proteomic data, high-resolution image data, high-frequency financial data, functional and longitudinal data, and network data, among others. Simultaneous variable selection and estimation is one of the key statistical problems in analyzing such complex data. There have been many advances on the variable selection problem for linear and generalized linear regression models in the past decades. However, more recently, regularization, or penalized, methods are becoming increasingly popular and many new developments have been established. The performance of APE when compared to classical shrinkage estimation strategies. We will consider the case when a model is not fully sparse. The parameter vector may consist of three parts: strong, weak, and zero coefficients. The relatively aggressive variable selection methods are often ignoring covariates with weak coefficients completely.