基于最优权值的选择性神经网络集成方法

SELECTIVE NEURAL NETWORK ENSEMBLE BASED ON OPTIMAL WEIGHTS

本文提出一种基于最优权值的选择性神经网络集成构造方法,在训练出个体神经网络之后,使用遗传算法计算出这些网络在加权平均方法中对应的最优权值,然后选择权值大于一定阈值的部分网络使用简单平均方法组成神经网络集成,理论分析和实验结果表明,与传统方法相比,本文方法使用部分网络能够取得更好的效果。

In this paper, a selective constructing approach for neural network ensemble based on optimal weights is proposed. This approach uses genetic algorithm to calculate the optimal weight values for the networks when the weighted average method is ased , and then those networks with weight larger than a given threshold value are ensembled using simple average. Theoretical analysis and experimental results show that this approach outperforms traditional ones.