Abstract:

The extreme learning machine (ELM) achieves good performance for classification and runs at a fast learning speed because of choosing the learning parameters of hidden nodes randomly. However, when the parameters of the hidden nodes are absolutely randomly chosen, the performance of ELM is not always optimal. The multiple hidden layer output matrices extreme learning machine (M-ELM) is proposed which optimizes the architecture of hidden nodes by weighted calculation of different output matrices, and the matrices weights and the output weights are analytically determined simultaneously. In addition, the feature level fusion of ELM can be achieved by this method. For the real word classification problems, simulation experiments verify that M-ELM can provide a better performance than ELM.