基于贝叶斯算法的神经网络优化方法

Optimization Based on Bayesian Neural Network Algorithm

提出了一种近似建模的前馈网络训练算法-贝叶斯算法,该方法能对模型中的未知量构造其后验分布,提高网络的泛化性能,获取对应于后验分布最大值的权值向量.结果表明,贝叶斯算法所建立的神经网络近似模型具有更高、更稳定的精度.

This paper presents a model for the approximation of the feed -forward network training algorithms： Bayesian algorithm. This methed can model the amount of construction unknown posterior distribution, improve the network generalization performance, access to the posterior distribution corresponding to the maximum weight vector. The results show the approximate model to be established on Bayesian neural network algorithms has a higher and more stable accuracy.