论文部分内容阅读
在前人关于动态结构神经网络研究的基础上，提出了一种综合性质的隐节点增删算法：由训练过程的均方差和误差衰减率来确定神经元的增删时刻，并采用矩阵分析的方法研究隐节点输出间的线性相关性，动态删除多余的隐节点，计算机仿真结果表明，采用该方法动态增删隐节点是有效的。
Based on the previous research on dynamic structure neural network, this paper proposes a new algorithm of adding or deleting hidden neurons by using the mean square error of training process and the rate of error decay, and uses the method of matrix analysis Hidden node output linear correlation, the dynamic deletion of extra hidden nodes, computer simulation results show that the use of this method to dynamically add or remove hidden nodes is effective.