基于单隐层复值神经网络的分类问题研究
首发时间:2017-05-27
摘要:本文提出单隐层复值神经网络(Complex-Valued Neural Network, CVNN),并对网络结构,参数和算法进行了介绍,对输入数据,网络权值和偏置项参数进行复数转换。采用适当的误差函数,利用梯度下降法对网络输出误差进行反向传播,最后的训练结果显示单隐层复值神经网络在两分类问题上,相比于实数BP神经网络,无论在收敛速度,训练误差还是测试精度上均有较高提升。在多分类问题上,单隐层复值神经网络在测试数据时也具有较高的测试精度。
For information in English, please click here
Research on Classification Problem Based on Single Hidden Layer Complex-Valued Neural Network
Abstract:This paper proposed a single hidden layer complex-valued neural network (CVNN), the network structure, parameters and algorithms are introduced, we converted the input data, network weights and bias parameters to the complex numbers. Using the appropriate error function, the gradient-based Back Propagation (BP) algorithm was derived from the output error of the network. The results showed that the single hidden layer complex-valued neural network on the binary classification problems, compared to the real BP neural network in convergence speed, training error and testing accuracy all have a higher performance. On the multi-classification problems, single hidden layer complex-valued neural network to the test data also has the high testing precision.
Keywords: complex-valued neural network gradient descent method Back Propagation algorithm classification
基金:
论文图表:
引用
No.4735485119621814****
同行评议
共计0人参与
勘误表
基于单隐层复值神经网络的分类问题研究
评论
全部评论0/1000