论文部分内容阅读
提出了一种基于独立元分析(ICA)方法的权值初始化方法和动态调整S型激励函数的斜率相结合的神经网络学习算法。该方法利用ICA从输入数据中提取显著的特征信息来初始化输入层到隐含层权值。而且通过使神经网络的输出位于激励函数的活动区域,对隐含层到输出层的权值进行初始化。在学习过程中,再对每个隐单元和输出单元的激励函数的斜率进行自动调整。最后通过计算机仿真实际的基准问题,验证了论文提出的方法的有效性。实验结果表明,所提出的方法能有效地加快多层前向神经网络的训练过程。
A neural network learning algorithm based on independent element analysis (ICA) weight initialization method and dynamic adjustment S-type excitation function’s slope is proposed. This method uses ICA to extract significant feature information from the input data to initialize the input layer to the hidden layer weights. And by making the output of neural network located in the active region of the excitation function, the weights of the hidden layer to the output layer are initialized. During the learning process, the slope of the excitation function for each hidden unit and output unit is automatically adjusted. Finally, the computer simulation of the actual benchmark problem, verify the effectiveness of the proposed method. Experimental results show that the proposed method can effectively speed up the training process of multi-layer neural network.