快速自顶向下优化神经网络结构的方法  

Fast Approach for Optimal Brain Surgeon

作　　者：杨钟瑾[1] 史忠科[1] 

机构地区：[1]西北工业大学,西安710072

出　　处：《系统仿真学报》2005年第1期162-165,共4页Journal of System Simulation

摘　　要：引荐了一种优化神经网络结构的改进方法。近年来,各种神经网络结构优化方法相继被提出,但素负盛名的还是神经中枢手术优化算法(OBS)。提出一种与修剪技术相关的快速方法,这种方法起始于适当复杂的初始网络,然后剔除冗余的神经元。研究说明,这种快速方法与以往的常规神经中枢手术优化算法相比,具有更好的归纳性能、更简单的网络结构和更快的学习速度。An improved pruning algorithm is presented. Although various pruning algorithms have been developed, possibly the most popular of these is optimal brain surgeon (OBS). The way used by the improved algorithm is to train an initial network that is reasonable large and then eliminate the unneeded parts. Experiment results of comparative studies with formal OBS show that the improved pruning algorithm provides improvement of generalization capability, and reduction of both neural network complexity and training time.

关 键 词：神经中枢手术优化算法 修剪 结构优化 自顶向下 反向传播算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]