Improved extreme learning machine for function approximation by encoding a priori information

In this letter, a class of improved extreme learning machines (ELM) encoding a priori information is proposed to obtain better generalization performance and much faster convergence rate for function approximation. According to Fourier series expansion theory, the hidden neurons activation functions in the improved ELM are sine and cosine functions. In addition, the improved ELM analytically determines the output weights of neural networks. Finally, experimental results are given to verify the efficiency and effectiveness of the improved ELM.