论文部分内容阅读
提出一种有效的U-D分解DFP和BFGS算法.该算法解决了H阵的正定性问题,保证了算法的数值稳定性,并大大提高了计算效率.对H阵的计算量分析表明,该算法的计算效率比普通方法高20%,比普通平方根方法高0.4n(n为H阵维数)倍.神经网络训练的应用表明,新算法比普通DPP和BFGS方法更有效、更准确.
An efficient U-D decomposition algorithm for DFP and BFGS is proposed. The algorithm solves the positive definite problem of H matrix, guarantees the numerical stability of the algorithm and greatly improves the computational efficiency. The computation of H matrix shows that the computational efficiency of the algorithm is 20% higher than that of the conventional method and 0.4n higher than the ordinary square root method (n is the dimension of H matrix). The application of neural network training shows that the new algorithm is more effective and accurate than the ordinary DPP and BFGS methods.