Extreme learning machine (ELM), proposed by Huang et al., has been shown apromising learning algorithm for single-hidden layer feedforward neuralnetworks (SLFNs). Nevertheless, because of the random choice of input weightsand biases, the ELM algorithm sometimes makes the hidden layer output matrix Hof SLFN not full column rank, which lowers the effectiveness of ELM. This paperdiscusses the effectiveness of ELM and proposes an improved algorithm calledEELM that makes a proper selection of the input weights and bias beforecalculating the output weights, which ensures the full column rank of H intheory. This improves to some extend the learning rate (testing accuracy,prediction accuracy, learning time) and the robustness property of thenetworks. The experimental results based on both the benchmark functionapproximation and real-world problems including classification and regressionapplications show the good performances of EELM.
展开▼