前馈神经网络的容错性研究

以前馈神经网络为研究对象，提出了一种容错型神经网络学习算法．将系统运行过程中可能发生的各类故障随机地引入网络训练过程，使系统获得更加稳健的内部表示．仿真结果表明，该学习算法能够有效地提高神经网络的容错能力和泛化能力．

On the Fault Tolerance of Feedforward Neural Networks

A fault tolerance neural networks learning algorithm is proposed based on feedforward neural networks. The failures occur during system running are introduced randomly when network is trained. The result shows that a more robust internal representation is acquired by the system. The simulation result shows that this algorithm enhances the fault tolerant ability and the generalization of neural networks efficiently.