摘要
本文讨论了一类连续Hopfield型神经网络绝对稳定的充分必要条件：若连接权矩阵是对称的，则保证系统绝对稳定的充分必要条件是连接权矩阵半负定（-T∈P0）：若连接权矩阵对角以外的元素均为非负的，则保证系统绝对稳定的充分必要条件是-T的所有主子行列式非负（-T∈K0）．此结论的意义在于给出的这类神经网络是在求解最优化问题时保证不出现伪响应的最大一类网络．
The necessary and sufficient condition for absolute stability of a class of continuous Hopfield neural networks are derive. When connection matrix is a symmetric matrix, the network system is absolutely stable (ABST)if and only if T is negative semidefine(-T∈P0 ), and When the off-diagonal entries of the connection matrix are nonnegative, the system is ABST if and only if all the principal minors of matrix-T are nonnegative (- T∈K0 ). The most significant theoretical implication is that this type of neural network is the largest class of network that can be employed for solving optimization problems without spurious responses.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1998年第8期32-36,共5页
Acta Electronica Sinica
基金
国家教委博士点基金
关键词
神经网络
绝对稳定性
充分必要条件
优化问题
Neural network, Absolute stability, Necessary and sufficient condition, Optimization problems
