Efficient Method for Designing Associative Memory with Contextual Small-World Architecture

Jing Yang, Lixin He, Bin Kong

Abstract: The micro-structure of a neural network is known to have great influence on its macro-dynamic behaviors. For sake of imitating the inherent characteristics of small-world property discovered in biological neural system and human intelligence of associative memory via integrating geometrical structure information, an efficient method for designing associative memory neural network with contextual small-world architecture was proposed. Energy function stability demonstration and leaning rule was presented. Connectivity of the network was constructed to reflect contextual local spatial continuity of neuron connections. Numerical experiments demonstrate the effectiveness of this designing method in terms of memory and recall abilities enhancement.