论文部分内容阅读
剪枝算法是一种通过简化网络结构来避免过拟合的有效方法之一。文章依据Shannon熵原理定义了神经网络隐层节点输出的拟熵,该熵与Shannon熵对不确定性的描述具有相同的效果,但克服了Shannon熵中无定义和零值的缺点。将交叉熵和隐节点输出拟熵作为目标函数,并采用熵周期的策略对网络参数进行寻优,通过删除合并隐层神经元达到简化网络结构的目的。仿真结果表明,此方法简单易行,对BP网络的泛化性能有较好的改善。
Pruning algorithm is an effective way to avoid over-fitting by simplifying the network structure. According to the Shannon entropy principle, the article defines the pseudo entropy of the hidden layer node output of neural network, which has the same effect as the Shannon entropy to describe the uncertainty, but overcomes the shortcoming of no definition and zero value in Shannon entropy. Cross-entropy and pseudo-entropy of hidden node output are taken as objective function, and the entropy cycle strategy is used to optimize the network parameters. The network structure is simplified by deleting the hidden neurons. Simulation results show that this method is simple and easy to implement, and has a good improvement on the generalization performance of BP network.