Abstract: A method of Artificial Immune Network based neural network ENsemble(AINEN) is proposed. After initialing a network ensemble with Bagging, the artificial immune network is applied to the neural network ensemble which looks like an independent neural network in the microscopic view and an artificial immune network in the macroscopic view. Then the heterogeneity at microscopic level, and the fitness at macroscopic level is increased. As a result, the generalization error is decreased. The comparison of the results of AINEN and GASEN shows that the AINEN is more efficient than the GASEN.

Key words: artificial immune network, neural network ensemble, clone, mutation

摘要： 提出基于人工免疫网络的神经网络集成方法AINEN。在用Bagging生成神经网络集成之后，将人工免疫网络的原理应用到神经网络集成，组成了一个从微观上看是一个一个的神经网络，而从宏观上看是一个大的人工免疫网络的集成。通过在微观层次上提高神经网络集成的个体之间的异构度，在宏观层次上提高免疫网络的适应度，从而降低集成的泛化误差。AINEN与GASEN方法在标准数据集上进行的实验表明，AINEN能取得更小的泛化误差。

关键词: 人工免疫网络, 神经网络集成, 克隆, 变异

CLC Number: 

• TP18