Kernel Extreme Learning Machine with Mixture Correntropy for Face Recognition

Abstract

In this article, a robust kernel extreme learning machine (KELM) framework is designed using mixture correntropy for recognition of facial images. KELM is augmentation of ELM with kernel learning concept, has attained excellent performance in acknowledging numerous classification and regression problems. Due to random projection technique and no requisite of number of hidden neurons specified beforehand, KELM achieves better generalization performance than ELM. Since KELM is designed on MSE paradigm for Gaussian theory of noise, its efficiency may decrease for non-Gaussian cases. To enhance the learning speed and robustness of KELM, this work develops a new KELM framework with mixture correntropy (KELM-MXC) that opts mixture correntropy (MXC) as the optimization paradigm, instead of MSE. Experiments on face databases are performed to prove the effectiveness and comparison analogy with related state-of-the-art algorithms are reported to demonstrate the performance excellence of the presented algorithm.