Study Of The BAM Network Based On The Small Word Architecture

Artificial neural network simulate similar to our human brain, it has associative memory and a new computational structures. It can get the full information from a part of the given information and can imitate the human brain processing broken, inaccurate, or even very fuzzy information. Now, artificial neural network research has become a hot and important point.Firstly, the thesis introduces the research progress about associative memory and several associative memory model specifically the traditional all interconnected Hopfield network. The thesis points out the limitation of the Hopfield network, and then introduces the concept of BAM network; we give a detailed description about the BAM network's structure, forming principle, learning and memory rules. Then this thesis expounds the limitations of the BAM network model and its solution.Secondly, we introduce several sparse network structures, mainly including small-world networks and scale-free networks. We expound their algorithm, statistical properties etc. Aiming at the limitations of BAM network and shortcuts generation characteristics of small world network, we build a new BAM framework based on small world architecture.Finally, this thesis expounds the new BAM network model based on small world architecture.through introducing the concept of sparse connection into BAM network model, we simulate the new model on computer, through comparison the performances of the BAM system based on small word network and the traditional BAM network. We show that:the new BAM network model not only reserves the good fault-tolerance the traditional BAM network has,but also greatly improves the storage capacity, reduces calculation storage costs, and the cost of circuit implementation. This process is very helpful for the VLSL realization.