CAJ | 학술논문

为了提高极端学习机算法的稳定性和学习速度，结合L1／2正则化理论提出一种改进的极端学习机算法———基于L1／2正则化的快速学习算法（L1／2‐RELM ）。该算法首先采用L1／2正则项对极端学习机算法进行约束，其次运用half 算法确定网络输出权重，提高了算法的稳定性和学习速度。数值实验表明，所提算法的学习速度比极端学习机算法的学习速度更快，且性能更加稳定。
위료제고겁단학습궤산법적은정성화학습속도，결합L1／2정칙화이론제출일충개진적겁단학습궤산법———기우L1／2정칙화적쾌속학습산법（L1／2‐RELM ）。해산법수선채용L1／2정칙항대겁단학습궤산법진행약속，기차운용half 산법학정망락수출권중，제고료산법적은정성화학습속도。수치실험표명，소제산법적학습속도비겁단학습궤산법적학습속도경쾌，차성능경가은정。
In order to improve the stability and learning speed of extreme learning machine ,an improved extreme learning machine algorithm——— L1/2‐RELM is proposed by incorporating the theory of L1/2 regularization into ELM .In the proposed algorithm ,at first the regularization item is introduced to regularize the model of ELM al‐gorithm ,and then determining the output weights by using half algorithm which improves the stability and learning speed of the proposed algorithm .The simulation results show that the proposed algorithm has the faster learning speed and more stable performance than extreme learning machine algorithm .