An enhancing multiple kernel extreme learning machine based on deep learning

Kernel extreme learning machine(KELM) embeds raw data into high-dimensional feature space through predefined kernel functions to analyze and process data and has advantages in learning speed and generalization ability. Deep neural network(DNN) provide more effective data mapping capabilities than shallow kernel functions. In this article, we introduce a universal algorithm that combines DNN and KELM, named Multi Deep Kernel Extreme Learning Machine(MDKELM). It applies multi kernel learning methods to combine the mapping matrices of different layers of DNN and applies the obtained optimal kernel matrix to KELM. The experimental results on multiple benchmark datasets demonstrate the advantages and potential of this method.