A Detailed Analysis on Extreme Learning Machine and Novel Approaches Based on ELM
Ömer Faruk Ertuğrul, Yılmaz Kaya · 2015
Extreme learning machine (ELM) is a train method for single hidden layer feed forward neural network. The input weights and biases of ELM are selected randomly and output weights are determined analytically therefore ELM has a fast train stage. Unfortunately randomly selection of input weights and biases causes unstable accurate results. Accuracy of randomly selected input weights and biases (ELM) was compared with new proposed approaches: predefined input weights and biases (ELM-P) and determining input weights and biases by back propagation (ELM-B). Also novel approaches; single layer ELM (sELM), tuning ELM (tELM), ELM based on linear regression (ELMr) were proposed for determining the output weights instead of using Moore–Penrose generalized inverse method. The accuracies of proposed approaches were compared with each other and ELM. The results were showed that the proposed approaches are successful.