基于改进共轭梯度法的前馈网络快速监督学习算法

为了提高多层前馈神经网络的权参数的学习效率,通过引入改进的求解大规模线性方程组的共轭梯度法,提出一种新的基于LM的前馈网络学习算法.该算法不仅具有LM优化学习方法的快速收敛特性,而且降低了LM法的计算复杂度,可获得比其它标准算法更好的学习精度和推广预测能力.文中通过仿真结果证明了新算法在函数逼近和时间序列预测等问题环境下的有效性.

A Fast Supervised Learning Algorithm of Multilayer Feedforward Network Based on Improved Conjugate Gradient Method

To improve the weight learning efficiency of multilayer feedforward network,a new similar LM learning algorithm is proposed by introducing modified conjugate gradient method in solving of large-scale linear equation sets. In addition to the fast convergent advantage the LM method demonstrates that,the new algorithm not only reduces the training time and overall complexity, but also achieves training accuracy and generalization capability comparable to more standard approaches. Extensive simulation results are provided to show the effectiveness of the new algorithm.