关键词: 神经网络, 极限学习机, 最小二乘估计, 主成分估计

Abstract: Extreme Learning Machine（ELM） is a kind of Single-hidden Layer Feedforword Neural Networks（SLFN）. It is simpler in structure and faster with learning speed Comparing to traditional neural network algorithms. The output-weight of ELM is calculated by least square estimation. ELM may produce a poor and unreliable solution when the hidden layer output matrix is not full column rank, or when the training data is contaminated with outliers. Principal components estimation is an improved algorithm for least square estimation. The principal components estimation can significantly improve the robustness against data noise and outliers when the hidden layer outpu tmatrix is not full column rank. A novel approach based on principal components estimation of extreme learning machine called PC-ELM is derived in this paper. Simulation results indicate that the proposed algorithm in this paper can significantly improve the robustness against data noise and outliers as well as good generalization performance.

Key words: neural networks, extreme learning machine, least square estimation, principal components estimation