一类新的模式识别联想神经网络

本文提出一类可用于模式识别的联想神经网络的综合方法,这类网络结构不受对称联接的限制,网络保证了要求的M类模式的稳定形成,且网络的容量远远超过Hopfield的联想神经网络,网络渐近稳定平衡点的吸引特性使受噪声污染的模式能得以正确恢复,体现了神经网络的非线性滤波性质。文中给出了综合一个这类联想网络计算机模拟以及模式识别的例子。

A New Class of Associative Neural Networks in Pattern Recognition

In this paper, we give a new method for synthesis of a class of associ- ative neural networks used for pattern recognition. The associative memory networks synthesized here is not limited by asymmetric interconnection and can guarantce to be of M stable pattern formation desired. The capacity of the networks is over than that of Hopfield''s CAM. The properties of attraction domains of asymptotically stable equilibria in the network, representing a nonlincar filter characteristic, enable the pattern contaminated to be restored correctly. A network synthesized and pattern recongnition with computer simulation have been illustrated in the paper.