Face recognition based on kernelized extreme learning machine

Abstract

The original extreme learning machine (ELM), based on least square solutions, is an efficient learning algorithm used in "generalized" single-hidden layer feedforward networks (SLFNs) which need not be neuron alike. Latest development [1] shows that ELM can be implemented with kernels. Kernlized ELM can be seen as a variant of the conventional LS-SVM without the output bias b. In this paper, the performance comparison of LS-SVM and kernelized ELM is conducted over a benchmarking face recognition dataset. Simulation results show that the kernelized ELM outperforms LS-SVM in terms of both recognition prediction accuracy and training speed. © 2011 Springer-Verlag.