摘要
对离散霍普菲尔德神经网络（ＤＨＮＮ）的性质进行了分析．指出：这种网络在解决二次优化问题时不能保证收敛，而且即使收敛也不能保证全局最优，不适合于解决优化问题．本文提出了一种更有效的选代算法，其结构与ＤＨＮＮ相同，可用高度并行方式实现．通过简单地选择收敛系数可以使算法收敛于全局最优．由此算法还可导出高阶收敛算法．因此，该算法是实时实现二次优化的一条有效途径．
The convergence property of discrete Hopfield neural networks(DHN-N) is studied. It is pointed out that DHN'N is either not convergent or unable to get the global optical solution. A more effidrit algorithm is then proposed, that is suitable for massively paraflel computation since it has the same structure as DHNN. The convergenee of the new algorithm can be guarantee, and an algorithm with high orderof convergence can be derived. Thus, the new algorithm is an effident vary for afl theoptition of quadrahc problems.
出处
《北京理工大学学报》
EI
CAS
CSCD
1994年第1期1-5,共5页
Transactions of Beijing Institute of Technology
基金
国家教委优秀年轻教师基金
关键词
二次规划
神经网络
二次优化
quadratic programming, parallel/neural networks, high order convergence
