DocumentCode :
553093
Title :
Adaptation of Gaussian ARD kernel for multiclass classification
Author :
Tinghua Wang
Author_Institution :
Sch. of Math. & Comput. Sci., Gannan Normal Univ., Ganzhou, China
Volume :
2
fYear :
2011
fDate :
26-28 July 2011
Firstpage :
983
Lastpage :
986
Abstract :
The problem of optimizing Gaussian Automatic Relevance (ARD) kernel in a multiclass setting is considered. Unlike the conventional Gaussian kernel with a single width parameter, the Gaussian ARD kernel adopts multiple widths corresponding to the input features. We first present a model selection criterion named kernel distance-based class separability (KDCS) to evaluate the goodness of a kernel in multiclass classification scenario, then propose a gradient-based optimization algorithm to tune the width parameters of Gaussian ARD kernel via maximizing the KDCS criterion. This method is essentially a feature weighting method since each learned parameter indicates the relative importance of the corresponding feature. The proposed method is demonstrated with some UCI machine learning benchmark examples.
Keywords :
gradient methods; learning (artificial intelligence); optimisation; pattern classification; support vector machines; Gaussian ARD kernel adaptation; Gaussian automatic relevance kernel optimization; KDCS criterion maximization; UCI machine learning; feature weighting method; gradient-based optimization algorithm; kernel distance-based class separability; model selection criterion; multiclass classification; width parameters; Ionosphere; Kernel; Machine learning; Optimization; Support vector machine classification; Training; auto relevance determination (ARD); feature weighting; model selection; multiclass classification; support vector machines (SVMs);
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Fuzzy Systems and Knowledge Discovery (FSKD), 2011 Eighth International Conference on
Conference_Location :
Shanghai
Print_ISBN :
978-1-61284-180-9
Type :
conf
DOI :
10.1109/FSKD.2011.6019661
Filename :
6019661
Link To Document :
بازگشت