前馈神经网络以概率1全局收敛的新型混合学习算法NHLA

A New Hybrid Learning Algorithm Which Globally Convergeswith Probability 1 for Feedforward Neural Networks

首先提出变形的ＢＰ算法ＩＢＰ，然后将它与Ｓｏｌｉｓ和Ｗｅｔｓ的随机优化算法相结合，提出了新型混合学习算法ＮＨＬＡ。算法具有能以概率１全局收敛于误差函数最小值的性质。

A new hybrid learning algorithm NHLA, which combines the variant IBP of algorithm BP and the random optimization algorithm presented by Dr Solis and Wets, is presented. NHLA can ensure global convergence to the minimum of the error function with probability 1. Therefore it overcome the drawback of algorithm BP that it sometimes fails to converge to the global minimum of the error function. 