Kernel Machine Methods with High-Throughput Data

• Activity Number: 538
• Type: Invited
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: International Indian Statistical Association
• Author(s): Debashis Ghosh*+ and Xiang Zhan and Wen-Yu Hua
• Companies: Penn State and Penn State and Penn State

Abstract

A powerful technique from the data mining community that has begun to see widespread applications is the use of kernel machine methods. A major advance in this area has been embedding these methods into a statistically simple framework that allows for easy estimation and inference. In this talk, we provide a survey of various applications in which kernel machines have been developed. Statistical issues of estimation and testing will also be described. In addition, we will also describe equivalences with other statistical methods, and in particular with the class of distance covariance methods that have been developed by Szekely, Rizzo and collaborators. Construction of kernels that accommodate the particular type of data will also be described. Real-data examples will be used throughout the talk. Time permitting, we will discuss some technical results describing the approximation properties of kernel machines.