一种大样本的鲁棒光滑前向网络训练算法

A KIND OF THE FEEDFORWARD AND BIG SAMPLES ALGORITHM WITH ROBUSTNESS AND SMOOTHNESS

将置信区间和加权因子引入神经元网络能量函数，提高了网络训练样本的可靠性，增强了系统的鲁棒性，同时，考虑网络逼近函数的光滑特性，在能量函数中曲率对函数逼近的影响，提出了一种大样本训练的鲁棒光滑前向神经元网络训练算法，并对算法进行了改进，使之对函数的逼近不仅具有一定的光滑特性，而且特别适宜于大样本场合．仿真实验证明了算法的有效性．

In this paper the weight factors and the confidence interval are introduced into the energy function of the neural networks, the reliability and robustness for training samples are improved. In the same way, the smoothing property of approaching function is considered, the curvature is added in the energy function, the feedforward algorithm with robustness and smoothness for big samples is proposed. The improvements are followed, the approaching function has the smoothness and the algorithm fits for the occasion of a great of samples, the simulation results prove the effectiveness of the algorithm.