Computer Science and Engineering, Department of

 

Date of this Version

7-29-2004

Comments

University of Nebraska–Lincoln, Computer Science and Engineering
Technical Report TR-UNL-CSE-2004-0012
Issued 7/29/2004

Abstract

A variant of instance-based learning is described which detects periodic patterns in the presence of sparse data. A weighted average gives higher weight to values in the recent past as well as those at expected periods in the past. After each new measurement, the space of weights near the current set is searched for a set that minimizes the error of prediction, thus providing a learning mechanism. The method is described in terms of an application which minimizes file download times by choosing between available servers.

Share

COinS