外场作用下吸引子神经网中神经元活性的直接观察

本文对强稀疏联结吸引子神经网在持续外场作用下神经元的局域活性进行了考察.对恒稳定性的情形给出了其概率分布分式.数值计算对不同情形下其固定分布行为提供了直观而又生动的观察.将分布图与吸引子结构进行了有比较的分析.我们的结果表明:强外场将导致分布图上阻碍神经元正确驰豫过程的不利屏障峰的出现.利于理想检索的有效分布仅能由适度的外场来取得.而且适度的外场对分布或检索的改善又会因网络过载而受到限制.

DIRECT OBSERVATION OF NEURONAL ACTIVTTIES IN ATTRACTOR NETWORKS WITH EXTERNAL FIELDS

The local activities of neurons in the strongly diluted atteractor networks with persistent external fields are investigated in this paper. For constant stabilities, their probabilistic distribution formula is presented. Numerical calculations provide direct and lively observation of their fixed distribution behavior for different cases. The distribution figures are comparatively analyzed with the atteractor structure. Our results indicate that too strong an external field will cause the appearance of an unfavorable peak in the distribution picture which acts as a barrier to the correct relaxation neurons, and the effective distribution favorable for perfect retrieval can only be achieved by an intermadiate external field. Furthermore, the improvement of distribution or retrieval by an intermediate external field will get limited due to the overloading of the networks.