一种提高推广性能的复合神经网络模型

A Model of combinatorial Neural Network Which Improves Generalization Ability

本文利用前向人工神经网络推广性能与初始权值的关系，提出了一种新颖的复会神经网络模型一选举模型复合网络，用于提高前向网络的推广性能。本文还给出了预测该复合推广性能的理论计算公式，并进行了严格的数学证明。理论分析和仿真实验表明，选举模型复合网络是提高前向网络用于目标分类的推广性能之有效手段。

By using the relationship between generalization ability and initial weghts of a feedforward artificaialneural network(FFANN) a new model of combinatiorial neural network-The Voting Model of combinatiorialNetwork had been proposed in this paper to improve its generalization ability.Furthermore,a theoretical formulahas been given to predict generalization abliry of the combinatoral network and proved correct in mathematicalways.Both theoretical analysis and simulation results show that for target classfication it is an effective method toimprove FPANN's generalization ability by using The Voting Model of Combinatorial Network.