Design Of Learning Algorithms For Fuzzy Neural Networks

Artificial intelligence with uncertainty is a hot research area in artificial intelligence domain now.Fuzzy neural networks are important tools to construct nonlinear systems in uncertain knowledge environment.In the paper, the fuzzy associative memories and fuzzy Hopfield networks are researched deeply. The problems have been researched as follow: (1) Taking advantage of the concomitant implication operator of T_ex , which is a t-norm and was presented by Einstein, a simple efficient learning algorithm is proposed for the fuzzy associative memory based on fuzzy composition of Max and T_ex (Max- T_ex FAM). It is proved theoretically that, for arbitrary given training pattern pairs, if the Max- T_ex FAM has ability to store reliably them, then the proposed learning algorithm can find the maximum of all connected weight matrices which can ensure that the Max- T_ex FAM stores reliably these pattern pairs. Finally an experiment is given to illustrate the effectivity of the presented learning algorithms.(2)A novel learning algorithm of fuzzy Hopfield is proposed in this paper, which makes use of more correlation information than others. The information utilized includes the correlation among the elements of not only the sample patterns, but also the input pattern. The network could adjust the weight matrix dynamically according to the input pattern. The research in the paper improves the adaptability and the performance of the networks.