Study On Fuzzy Auto-associative Memory Neural Network Based On Small World Model

Associative Memory network is a special artificial neural network. It can obtain all the information from uncompleted and noisy information. Fuzzy associative comes from the combine of Fuzzy systems and neural networks, and widely used in artificial intelligence and pattern recognition. Most of Fuzzy associative memory networks are full connected, with the increase of number of neurons, the collection of neurons is increased too, and the hardware implementation is difficult. So the research of fuzzy associative memory structure becomes popular in recent years. at the same time, Biological study shows that the small world characteristics is prevalent in the biological brain. Therefore, from the point of biological, it is reasonable to apply the small world model to the fuzzy associative memory neural network.This paper's research is based on the structure of network, depth research on the generation algorithm of the small world architecture. Proposed a small network model generation improve algorithm based on the best selection of weight matrix, and verify the feasibility and effectiveness from experiment. Then apply to the Fuzzy Auto-associative memory network, form the Fuzzy associative memory network model based on small world model. Following ways embodies the research content and innovation:1.Introduces the concepts and research status of fuzzy associative memory network. Points out the problem of the leaning algorithm and hardware implementation of fuzzy associative memory neural network. Analyzes the feasibility of sparse connection of full connection fuzzy associative neural network.2.Research the small-world network which is the typical network model in the Complex network. The original small world architecture generation algorithm is randomness, lacks of certainty. Reference the Harmonious Unifying Hybrid Preferential Model and the optimal synaptic dilution strategy of complex dynamic network, proposed a generation algorithm with certainty.3.Combine the Small world theory and fuzzy associative memory neural network; train the full connection neural network with Max-T norm operator. Get the weights matrix. Guided by the weights matrix, generated the fuzzy associative memory neural network based on small world model. This model is certainty; it retains the useful edge and deletes the useless edges, so kept the best performance with less connected price.4.Apply the fuzzy associative memory neural network based on small world model to face recognition. The results show that it's feasible and effective, robustness on recovery of noisy information.