An Adaptive Non-symmetric Fuzzy Activation Function-Based Extreme Learning Machines for Face Recognition

Abstract

The present investigation describes an adaptive non-symmetric fuzzy activation function-based extreme learning machines (ANF-ELM) for face recognition. ELMs are biologically inspired single hidden layer feed forward networks which present significant advantages over traditional back-propagation algorithm. Advantages of ELMs are low computational cost, fast learning speed, ease to implement and good performance results. In ELM, the hidden layer parameters are randomly selected and then the output weights are determined by calculating the inverse of the outputs of hidden layer. ANF-ELM is a simple single hidden layer feed forward network which gives nonlinear mapping from the input space to feature space by an adaptive non-symmetric fuzzy activation function (FAF), s. The s FAF is non-symmetric in nature with shifted origin which gives better performance compared to the symmetric activation function. The effectiveness of the ANF-ELM classifier is tested on four datasets: AT&T, Yale faces, CMU PIE and UMIST. The results show that an ANF-ELM provides good performance results and faster training speed when compared to other state of the art techniques.