离散型Hopfield神经网络学习算法  

Learning algorithm for discrete Hopfield neural network

作　　者：朱丽萍[1,2] 王锋辉[1,2] 李洪奇 吕洁[1,2] SikandarAli 

机构地区：[1]中国石油大学(北京)石油数据挖掘北京市重点实验室,北京102249 [2]中国石油大学(北京)地球物理与信息工程学院,北京102249

出　　处：《计算机工程与设计》2018年第3期831-835,共5页Computer Engineering and Design

摘　　要：为使得离散型Hopfield神经网络(DHNN)具备更强的联想记忆功能,基于替换函数f(x),权值求取采用提出的学习算法。通过设计比sgn更强约束能力的函数f(x),在满足sgn函数要求的同时连续可导,由于f(x)连续可导,可根据能量最低点网络状态不再发生变化的特性定义损失函数,用梯度下降算法来求解。使用Matlab编程验证效果,验证结果表明,该学习算法比传统的外积法、正交设计法具有更好的效果,对原始信息还原率提高了5%-11%。To make the discrete Hopfield neural networks have stronger associative memory a b i l i ty, a learnweight of DHNN was adopted. Function /（.x ） was designed w i th stronger in h ib it io n than sgn fu n c t io n, which quirements of sgn function and possessed the continuously differentiable feature at the same /（x ） was continuously differentiable, according to the characteristics of the netw o rk that the lowest energy state doesn’ t changeanymore,the gradient descent algorithm was used to solve the problem. M a t la bw a s used to ve rify the effects. Results show that the performance of the proposed method is better than outer product method and other traditional methods and the original information reduction rate increases by 5% to 11%.

关 键 词：离散型Hopfeld神经网络 外积和法 损失函数 梯度下降算法 联想记忆 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]