| Paper: | MLSP-L2.6 | ||
| Session: | Kernel Machines | ||
| Time: | Wednesday, May 17, 18:10 - 18:30 | ||
| Presentation: | Lecture | ||
| Topic: | Machine Learning for Signal Processing: Signal detection, Pattern Recognition and Classification | ||
| Title: | Controlling False Alarms with Support Vector Machines | ||
| Authors: | Mark Davenport, Richard Baraniuk, Clayton Scott, Rice University, United States | ||
| Abstract: | We study the problem of designing support vector classifiers with respect to a Neyman-Pearson criterion. Specifically, given a user-specified level $\alpha \in (0,1)$, how can we ensure a false alarm rate no greater than $\alpha$ while minimizing the miss rate? We examine two approaches, one based on shifting the offset of a conventionally trained SVM and the other based on the introduction of class-specific weights. Our contributions include a novel heuristic for improved error estimation and a strategy for efficiently searching the parameter space of the second method. We also provide a characterization of the feasible parameter set of the $2\nu$-SVM on which the second approach is based. The proposed methods are compared on four benchmark datasets. | ||
©2018 Conference Management Services, Inc. -||- email: webmaster@icassp2006.org -||- Last updated Friday, August 17, 2012