论文部分内容阅读
给出了一种优化三层神经网络结构算法。首先以较大隐层节点数进行学习,然后根据隐层输出信息提取各节点之间的线性特征来优化隐层节点数。对隐层输出信息提取各节点之间的线性特征,给出了两种方法:一种是在BP神经网络迭代后用自适应线性单元来提取隐层输出各节点之间线性特征,另一种是在BP神经网络迭代时就尽量使隐层输出各节点之间呈线性,然后用上种方法来提取隐层输出各节点之间线性特征。实例验证,后一种比前一种能更好地优化BP神经网络结构。
An optimization algorithm of three-layer neural network structure is given. First, we study the number of hidden nodes, and then optimize the hidden nodes by extracting the linear features of each node according to the hidden layer output information. Two methods are proposed to extract the linear features of each node from the hidden layer output information: one is to extract the linear features of hidden nodes output nodes by using adaptive linear elements after iteration of BP neural network, and the other In the iteration of BP neural network, the output of hidden layer is made as linear as possible between nodes, then the method of extracting the linear features of hidden nodes output nodes. The example verifies that the latter one can better optimize BP neural network structure than the former one.