|
Activity Number:
|
191
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Physical and Engineering Sciences
|
| Abstract - #300922 |
|
Title:
|
Divergence-Based Kernel for Spectrum Classification and Its Applications
|
|
Author(s):
|
Tsukasa Ishigaki*+ and Tomoyuki Higuchi
|
|
Companies:
|
The Institute of Statistical Mathematics and The Institute of Statistical Mathematics
|
|
Address:
|
4-6-7 Minami-Azabu, Tokyo, 106-8569, Japan
|
|
Keywords:
|
Divergence-based kernel ; classification ; kernel method ; kernel classifier ; acoustic signal
|
|
Abstract:
|
The proper selection of the kernel function is important in order to obtain the high classification performance for the classifier with kernel method. The presentation describes a high-accuracy classification for the spectra using the kernel classifiers (Support Vector Machine and Relevance Vector Machine) with the divergence-based kernels. The proposed method introduces the divergence, which is a metric between two probability distributions, as a kernel function for similarity calculation of two spectra with the appropriate statistical signal processing. We apply the method to three kinds of acoustic signals observed from real systems. The proposed method demonstrates a higher accuracy than popular kernels, such as the polynomial and Gaussian kernels.
|
