A Factorized Extreme Learning Machine and Its Applications in EEG-Based Emotion Recognition

Extreme learning machine (ELM) is an efficient learning algorithm for single hidden layer feed forward neural networks. Its main feature is the random generation of the hidden layer weights and biases and then we only need to determine the output weights in model learning. However, the random mapping in ELM impairs the discriminative information of data to certain extent, which brings side effects for the output weight matrix to well capture the essential data properties. In this paper, we propose a factorized extreme learning machine (FELM) by incorporating another hidden layer between the ELM hidden layer and the output layer. Mathematically, the original output matrix is factorized so as to effectively explore the structured discriminative information of data. That is, we constrain the group sparsity of data representation in the new hidden layer, which will be further projected to the output layer. An efficient learning algorithm is proposed to optimize the objective of the proposed FELM model. Extensive experiments on EEG-based emotion recognition show the effectiveness of FELM.