Machine learning for signal processing: kernel-extreme learning machines with augmented complex-valued and mixture correntropy techniques

In this paper, we propose novel machine learning algorithms for signal processing applications based on kernel-extreme learning machines (KELM) and its variants. Specifically, we focus on the use of complex domain and mixture correntropy techniques to enhance the performance and generalization ability of KELM. Two models are derived by augmenting a complex-valued extreme learning machine with kernel functions, minimizing human intervention while enhancing accuracy. Furthermore, we extend the mixture correntropy-based kernel extreme learning machine (MCKELM) to complex domains, resulting in the MCKCELM algorithm, capable of solving prediction problems in complex domains. To further improve the performance of MCKCELM, we apply an augmented matrix to derive two variant models, and demonstrate their convergence through experiments. All models presented in this paper are regularized models specifically designed for signal processing tasks.