Day & Time
25th July, 2014 15:10 ~
Room C816, Building of the Faculty of Science Graduate school of Science
Optimum Nonlinear Discriminant Analysis and Discriminant Kernels
Recently the kernel discriminant analysis has been successfully applied in many applications. However, the kernel function is usually defined a priori and it is not known what is the optimum kernel function for nonlinear discriminant analysis. Also the class information is not usually introduced to define the kernel functions. In this talk, the optimal kernel function in terms of the discriminant criterion (called discriminant kernel) is shown by investigating the optimum discriminant mapping constructed by optimum nonlinear discriminant analysis. The discriminant kernel is given by using the Bayesian a posterior probabilities. For real applications, a family of discriminant kernels can be derived by changing the estimation method of the Bayesian a posterior probabilities. Some examples of discriminant kernels are shown and the effectiveness of discriminant kernels is verified by several experiments using UCI ML repository datasets.