Output details

A signal theory approach to support vector classification: the sinc kernel

Significance of output:

<24>Support Vector Classification has rightly become a popular technique in artificial intelligence. This paper addresses an important outstanding challenge of how to design an appropriate kernel function by making a theoretical connection with ideas from signal theory. The connection is novel and the theory provides a principled means for determining a restricted finite search space wherein the optimal kernel parameterisation lies. This is advantageous in terms of computational efficiency and generalisation performance. This paper presents the mathematical theory of the approach and validates it through application to existing hyperspectral image data.