Kernel Extreme Learning Machine Based On K Interpolation Simplex Method

The kernel extreme learning machine eliminates the morbid problem of the coefficient matrix of the traditional limit learning machine,However,the kernel function used in its training model has some problems such as the type of kernel function and the selection of kernel parameters,which will lead to the slow convergence of the training of the kernel extreme learning machine and the decline of the classification precision.To solve the above problems,a K interpolation simplex kernel extreme learning machine is proposed in this paper.(1)In order to solve the problem that the kernel parameter of the Gauss kernel function is hard to be optimized,Nelder-Mead simplex method is used to search the optimal kernel parameters for the kernel extreme learning machine.The Nelder-Mead simplex method is used to search the optimal kernel parameters,and the K interpolation method is introduced to provide initial search values for Nelder-Mead simplex.The problem of Nelder-Mead simplex sensitivity to the initial value of kernel parameters is solved,and the simplex method is avoided to get into the local optimal solution.Thus,the kernel parameter adaptive optimization of kernel extreme learning machine is realized,and the classification accuracy of kernel extreme learning machine is improved.(2)In order to solve the problem that the kernel function selection for kernel extreme learning machines,the combined kernel function method using Gauss kernel and polynomial kernel is proposed to overcome the deficiency of single kernel function and exert the two kernel function nonlinear feature mapping ability.The main methods are as follows:first,the optimal kernel parameters are determined by the K interpolation simplex method.Then the optimal combination strategy is used,the combination coefficient is adjusted,and a new combinatorial kernel function is constructed to improve the identification ability of the nonlinear characteristics.Finally,the effectiveness of the algorithm is verified on the UCI dataset and the pattern recognition problem.The experimental results show that the classification accuracy of the proposed algorithm is better than that of the phase comparison algorithm,and the idea of improving the kernel extreme learning machine is feasible.