Abstract:

The principle of intelligent optimization algorithms is complex and the setting of their parameters is difficult，so it is difficult for intelligent optimization algorithms to optimize the parameters of kernel Fisher discriminant analysis(KFDA). A quasi-Newton algorithm to automatically optimize the multiple parameters of KFDA is proposed. The objective function is constructed using an empirical risk minimization principle. To make the objective function continuous and derivative, a sigmoid function is introduced to transform the discrete binary output of KFDA into continuous probability output. The initial parameters are selected by orthogonal array. Experimental results indicate that the classification performance of the proposed algorithm is close to the genetic algorithm. The higher convergence rate and simpler principle are obtained by using the proposed algorithm in comparison with the genetic algorithm. The proposed algorithm can be effectively used to optimize the multiple kernel parameters of KFDA.