图样间联想神经网络的稳定性研究

图样间联想神经网络作为新型的联想模型,其联想功能和光学实现较Hopfield等图样内联想网络都更具优越性;但至今缺乏对图样间联想神经网络系统性能的全面完备的认知。文章详细研究了图样间联想网络的动力学稳定性,论证了恒零阈值下网络单步稳定的特性;并根据稳定性分析的直接法引入代表系统矛盾程度的准能量函数考察了网络趋向稳定的方式,结合实验统计指出图样间联想网络一般阈值下是渐近稳定的。据此提出了模型的阈值改进方案,引入了反馈,使网络联想性能在原模型程度上有所提高。

Study on the Stability of Interpattern Association Neural Network

The newly advanced association model—— — interpattern association neural network(IPA)is superior to those of intrapattern associations such like the Hopfield model in both computing performances and optical implementations,while comprehensive cognition of IPA neural system are still to be known.In this paper,the neurodynamic stability of IPA model is studied in particular.It's proved that the net system comes to stable within one iteration in constant zero thresholds and is asymptotically stable in general instances according to Lyapunov's direct method and computer simu-lations.A threshold modification is then put forward to improve IPA's associating ability by introducing feedback to the network at the same time.