一种基于鲁棒估计的极限学习机方法

极限学习机(ELM)是一种单隐层前馈神经网络(single-hidden layer feedforward neural networks,SLFNs),它相较于传统神经网络算法来说结构简单,具有较快的学习速度和良好的泛化性能等优点。ELM的输出权值是由最小二乘法(least square,LE)计算得出,然而经典的LS估计的抗差能力较差,容易夸大离群点和噪声的影响,从而造成训练出的参数模型不准确甚至得到完全错误的结果。为了解决此问题,提出一种基于M估计的采用加权最小二乘方法来取代最小二乘法计算输出权值的鲁棒极限学习机算法(RBELM),通过对多个数据集进行回归和分类分析实验,结果表明,该方法能够有效降低异常值的影响,具有良好的抗差能力。

Extreme learning machine on robust estimation

Extreme learning machineELM is a kind of single-hidden layer feedforword neural networksSLFNs. Comparing with traditional neural network algorithms, it is simpler in structure, with higher learning speed, and good generalization performance. The output-weight of ELM was calculated by LSEleast square estimation method. However, LSE lack of robustness, the result would be seriously damaged when there were outliers in the training data. In order to solve this problem, this paper derived a novel approach based on M-estimators of extreme learning machine called RBELM. Simulation results indicate that the RBELM proposed can significantly robust against data noise and outliers.