Simulation Study Of Morphological Associative Memory For Implicit Learning

The research of implicit learning is related to the fundamental problems of human potential development and the major issues of human know itself, it has become a difficult and hot research in the cognitive psychology increasingly. Status of implicit learning is established, the researcher have taken a variety of methods to discuss more in-depth the psychological mechanism, essential characteristics, the relationship with traditional explicit learning an so on of implicit learning. Because artificial neural network model and the essential characteristics of implicit learning has the very similar match, so it is especially suitable for the research of implicit learning. For the two kind of more general implicit learning tasks-artificial grammar learning and sequence learning, the more widely used neural network model-auto associate and simple recurrent network have been used to simulate the implicit, in order to discuss the nature and mechanisms of it. However, so far, researcher are still used traditional artificial neural network to simulate implicit learning. Due to some defects of traditional artificial neural network itself, such as low efficiency of learning, limited storage capacity, and the convergence problem of weight adjustment requiring repeated loop iterations, there are many drawbacks in the process of these defects lead to implicit learning simulation. In this paper, the morphological associative memories(MAM) are used in order to solve problem.Firstly, for the implicit, which concept,the distinguish, connection and interactions between the traditional explicit learning have been summarized and analyzed. And the two kind of more general implicit learning tasks-artificial grammar learning and sequence learning be analyzed. On this basis, some of essential characteristics of the implicit are summarized.Secondly, for morphological associative memories this novel class of artificial neural networks, which storage performance, memory performance, and noise immunity are analyzed. Theoretical analysis shows that MAM with unlimited storage capacity, certain ability of resisting erosive noise and dilative noise, as well as completely recall under certain conditions, especially with one-shot recall, thus there is no convergence of traditional neural network. By using these advantages of MAM, analyze the applicability, authenticity, and efficiency of using MAM to simulate the implicit learning. At last, the experimental analysis shows that used MAM to simulate the implicit learning, there is great improvement of the authenticity and efficiency of simulation that compared with the artificial neural network. This will provides a new perspective and tools for implicit learning research, and also will expand the field of application of the MAM further.