Study On A Novel Associative Memory Model Based On Circuit Expression Of Information

Associative learning and memorzing is an important mechanism to realize the cognitive function and promote the emergence of intelligent behaviors. The problems of associating mechanism, memory capacity, neural expression of signal and information, rhythm and state shift of associative thinking, emergence of intelligence are all the current research hotspots in the field of cognitive science and intelligence science. To solve these problems, from the cognitive psychological level, the author studied the related theories such as network graph theory, artificial neural computation, circuit reverberating theory, associative memory, and then established a novel multi-functional associative memory model based on the mechanism of local dynamic searching for elastic network and the algorithm structure of multi-agent distributed evolutionary computation, to discuss the problems of pattern learning, pattern association, pattern relating and evolving dynamics, as well as the related algorithms, software designing and simulating. The main contributions are as follows:1. Based on the "circuit reverberating theory", the mechanism of local dynamic searching for elastic network and the algorithm structure of multi-agent distributed evolutionary computation, a novel multi- functional associative memory model was established. This model was characteristic of high memory capacity and strong association functions.2. Based on the established model, some associative pattern learning algorithms were put forward, which had the physiological features of structure's self-organizing and neural circuits' reverberating.3. The dynamics of pattern associating was analyzed. Associating behavior stimulated by external environment was equivalent to the effective searching under the restriction of partial pattern features' matching Associating behavior stimulated by internal environment had the thinking functions of multi-angle association and chain-typed association.4. Aiming at the association function, the optimizing algorithms of dynamic network local searching were designed based on parallel genetic algorithm framework, Some strategies such as distributed elastic resonating, self-adaptive elastic adjusting and occasional interaction between human and machine were adopted, which strengthened the system's maintenance and enhanced local searching, real-time processing and dynamic adaptation.