混沌搜索神经网络集成求解广义异或分类问题

Generalized XOR Classification Solution Using Neural Networks Ensemble Based on Chaos Searching

神经网络集成被认为是最有效的工程化神经网络设计方法之一.把混沌搜索技术引入到神经网络集成方法中,提出了基于混沌搜索的神经网络集成方法.该方法保持了神经网络集成可以有效提高神经网络的泛化能力的特点,减小"多维共线性"和样本噪声的影响.为有效保证网络权值多样性,在网络权值初始化的过程中加入混沌理论.对广义异或问题的仿真计算表明,该方法的性能优于标准的神经网络集成方法.

Neural network ensemble（NNE）is a most powerful method for design of neural network.This paper incorporates chaos searching technologies into neural network ensemble methods,and proposes a chaos-based NNE（CNNE）method.This proposed method maintains the characteristics that may enhance the generalization performance of neural networks and decrease the effects of multidimensional co-linearity and noise of sample.Chaos theory is adopted in the process of initialization of weight values in order to effectively guarantee the diversity of weight values of networks.The computational results on the generalized XOR problem show that the proposed method is superior to original neural network ensemble.