An improved Extreme Learning Machine

The input weight and the bias of hidden layer nodes are randomly generated in the training process of the traditional Extreme Learning Machine (ELM), which is simple and with no need for repeated iteration, thus the model training speed increases significantly. However, this algorithm model possesses a defect that it´s difficult to choose a reasonable network structure for users due to the parameters which are randomly generated each time can not guarantee high accuracy. For the requirements on rapidity and accuracy of ELM, an improved ELM model is proposed in this paper, the sine function data and its variants function data are adopted to be as the training sample and testing sample. The algorithm execution of generating two paremeters randomly has been circulated n times by defining cycle number n, and then a group of parameters with highest accuracy are automatically selected from the n cycles. Experimental results show that, compared with traditional ELM, the model accuracy of the proposed scheme improves greatly; then compared with BP neural network, the improved ELM model in this paper has great advantage in training speed, and also can achieve very high precision in accuracy.