Fuzzy Associative Memories Networks Based On Cloud Computing Platform

Fuzzy associative memory networks are important tools of uncertainty of artificial intelligence technology. It combines the advantages of the neural networks and fuzzy logic, bearing a lot of advantages in dealing with issues such as non-linear and fuzzy and having a huge potential in intelligent information processing. However, with the development of the Internet, the data showed the increasingly rapid growth trend, the face of the massive amount of data processing, many intelligent methods are faced with enormous challenges. The emergence of cloud computing provide this kind of problem with the new solution. Cloud computing can integrate a large number of cheap computing resources, with centralized management of computing facilities management, to provide computing services to the external. Google, Amazon, IBM, Microsoft, and Yahoo, the company is a pioneer in cloud computing. Hadoop is a software framework for distributed processing of the Apache open-source organization, with the function of the distribution of processing large amounts of data, applications running on a cluster can be composed of a large number of low-cost hardware, it implements Google’s MapReduce programming model, provide application with reliable interfaces. The goal is to build high reliability and good scalability of distributed systems. through parallel transformation, Many algorithms can run on the Hadoop platform, and then by running on a cluster equipped with Hadoop platform, thereby reducing the execution time, improve computational efficiency.The triangular norm fuzzy nature of associative memory networks and fuzzy associative memory network robustness defined, this paper analysed the global robustness of fuzzy bidirectional associative memory networks learning algorithm based on Einstein t-norm. When training mode perturbation is positive perturbation, the conclusion that the learning algorithm can maintain good robustness is proved in theory and in experimental validation, the dynamic relationship between the training mode perturbation and output mode perturbation based several operators are made contrast; when the perturbation contains negative numbers, the learning algorithm does not meet global robustness. What is further, the relationship between the max perturbations of training pattern set and output pattern set is studied, and the curve is gotten comparing with other operators. Then the parallelized learning algorithm of the fuzzy associative memories networks is researched and implemented on Hadoop cloud computation platform. By comparing the efficiency of processing large data on a single computer and cluster computers, the conclusion that after parallelizing, the time-shorting effect of the learning algorithm will be more obvious with the growth of the amount of data.