|
Activity Number:
|
251
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #309061 |
|
Title:
|
Algorithms for Support Vector Machines
|
|
Author(s):
|
Denise Reeves*+
|
|
Companies:
|
George Mason University
|
|
Address:
|
9984 Hemlock Woods Lane, Burke, VA, 22015,
|
|
Keywords:
|
ill-posed inverse problems ; bias and variance dilemma ; optimal coordinate system
|
|
Abstract:
|
Estimation of sufficient statistics and parameters are ill-posed inverse problems characterized by partially specified system models. Such problems require mathematical constructs that rule out physically implausible solutions with simple, eloquent representations. Classical estimation techniques fail to mitigate the bias and variance dilemma. The right structure or bias is often an optimal coordinate system. The likelihood ratio is generally reduced to a sufficient statistic. We consider the optimal curve and coordinate system being estimated by the support vector algorithm and precisely clarify the nature of the bias for common covariance matrices.
|
