神经网络在预测编码中的应用

讨论如何有效地利用backpropagation(BP)神经网络进行预测编码。改进的BP算法可以提高BP神经网络地收敛速度、泛化能力和稳定性。一般而言 ,四层网络比三层网络具有更快地学习速度 ;二阶网络比一阶网络性能更优越。利用MATLAB程序 ,构造三层、四层和二阶BP神经网络进行预测编码。测试结果表明 ,利用规模足够大的三层BP网络进行预测编码可以取得较好效果 ;四层和二阶网络比等价的三层网络可以更有效地进行预测编码

Application of the Neural Network in Predictive Coding

This paper discusses how to perform predictive coding effectively with BP neural network. The modified BP algorithm has a higher learning speed, better generalization ability and stability. In general, a four layer BP network has a faster learning speed than a three layer BP network, and a second BP network is better than a first order BP network. Three layer and four layer BP neural networks and second order BP neural network, which are constructed with MATLAB, are used to perform predictive coding. The test results show that the three layer BP neural network with enough size can be used to perform predictive coding efficiently and the second order BP neural network is mor powerful for predictive coding.