适于前馈神经网络的
LM-QuasiNewton综合学习新方法

New Comprehensive Learning Method Based on LM-QuasiNewton Algorithm Applying for Feed-Forward Neural Network

摘要: 为实时解决神经网络学习过程中可能遇到的大残量时的收敛问题,将LM算法与QuasiNewton优化算法结合,构建了一种综合学习算法(LM-QuasiNewton算法).仿真算例表明,该算法较好地解决了残量问题,收敛性与稳定性优于其它权值算法.合学习算法.仿真实例表明,该算法较好地解决了残量问题,在收敛性与稳定性方面优于其它权值算法.
- 前馈 /
- 神经网络 /
- 收敛 /
- QuasiNewton算法 /
- LM算法 /
- 残量问题
Abstract: To solve the large residual problems which may occur during weight training, QuasiNewton algorithm was combined with Levernberg-Marquardt algorithm to form a new comprehensive learning algorithm, named LM-QuasiNewton, for feed-forward neural networks. Simulation shows that LM- QuasiNewton algorithm can effectively solve large residual problems with better convergence and stability compared with otherweight learning algorithms.
- feed-forward /
- neural networks /
- convergence /
- QuasiNewton algorithm /
- Levernberg-Marquardt algorithm /
- residual problem

点击查看大图