摘要
借鉴细胞学中的干细胞理论,提出了基于干细胞机制的进化神经网络(SCEABPNN),以实现误差反向传播神经网络(BPNN)的优化。SCEABPNN不需要遗传神经网络(GABPNN)中的编码、解码、交叉、变异等操作,把网络中的结点看成细胞,通过细胞移植、细胞置换、细胞凋亡操作进行网络优化。仿真结果证实,SCEABPNN不但可以全局收敛,有效解决BPNN易陷局部最小的问题,而且收敛速度比GABPNN和标准BPNN更快。
A new algorithm,SCEABPNN for short,was proposed based on stem cell mechanism in cytology to optimize the back propagation neural network（BPNN）.SCEABPNN does not need such operations as encoding,decoding,crossover,mutation in genetic back propagation neural network（GABPNN）.It treats a node in a neural network as a cell and optimizes a neural network by such operations as cell transplantation,cell replacement and cell apoptosis and so on.Simulation results confirm that SCEABPNN can ensure a BPNN escape from local minimum to converge globally and its convergence speed is higher than GABPNN and standard BPNN.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2011年第12期2629-2634,2646,共7页
Journal of System Simulation
基金
国家重点基础研究发展计划(973计划)(2008CB317107)
关键词
BP神经网络
干细胞机制
优化算法
机器学习
back propagation neural network
stem cell mechanism
optimizing algorithm
machine learning
