A Novel Extreme Learning Machine Based on Hybrid Kernel Function

Abstract

Extreme learning machine is a new learning algorithm for the single hidden layer feedforward neural networks (SLFNs). ELM has been widely used in various fields and applications to overcome the slow training speed and over-fitting problems of the conventional neural network learning algorithms. ELM algorithm is based on the empirical risk minimization, without considering the structural risk and this may lead to over-fitting problems and at the same time, it is with poor controllability and robustness. For these deficiencies, an optimization method is proposed in this paper, a novel extreme learning machine based on hybrid kernel function (HKELM). The method constructs a hybrid kernel function with better performance by fully combining local kernel function strong learning ability and global kernel function strong generalization ability. Compared with traditional ELM, the results show that this method can effectively improve the ELM classification results, avoid local minimum, with better generalization, robustness, controllability and faster learning rate.