带偏置的选择性神经网络集成构造方法

训练多个神经网络并将其结果进行合成，能显著地提高神经网络系统的泛化能力。本文提出了一种带偏置的选择性神经网络集成构造方法。对个体网络引入偏置项，增加可选网络的数量。选择部分网络集成，改善网络集成的性能。把个体网络的偏置项统一为集成偏置项，在训练出个体神经网络后，使用遗传算法选择部分网络集成，同时确定集成偏置项。理论分析和实验结果表明，该方法能够取得很好的网络集成效果。

Selective Neural Network Ensemble with Bias

Neural network ensemble is a learning paradigm, which integrates many neural networks into an ensemble to solve a problem jointly. In this paper, the relationship between the ensemble and its component neural networks is analyzed and a novel approach for the construction of the selective neural network ensemble with bias is proposed. The genetic algorithm is used to select part of the trained individual networks to be integrated and the bias is deter- mined to correct the combination of the ensemble simultaneously. Theoreticalanalysis and experimental results show that this approach allows one to design effective neural network ensemble.