부산대학교 도서관

In the present paper, we investigate the dependence on the configuration of inhibited in-coming synaptic connections in Nara & Davis chaotic neural network model related with the sensitive response of the chaotic wandering state to memory pattern fragments. It has been shown that when a memory pattern fragment was given, the chaotic wandering state in Nara & Davis chaotic neural network model suddenly converges into the target memory pattern with a high robustness. However, the potentiality depends on dynamical properties of the chaotic wandering states, which are affected with the configuration of the inhibited in-coming synaptic connections. Therefore, we investigate the sensitivity to memory pattern fragments for two types of the configuration, (i) which denotes randomly inhibited in-coming synaptic connections except for ones from memory pattern fragments referred as a partly random configuration hereafter and (ii) which defines randomly inhibited in-coming synaptic connections referred as a fully random configuration hereafter. From the computer experiments, the success ratio for the partly random configuration becomes much higher than the fully random ones, and the accessing time becomes shorter. In addition, from Lyapunov dimension, the system with the partly random configuration reveals higher developed chaos.