Training T-S norm neural networks to refine weights for fuzzy if-then rules

This correspondence proposes an approach to learning weights of weighted fuzzy if-then rules. According to a given T-S norm-based reasoning mechanism, this approach first maps a set of weighted fuzzy if-then rules into a feed-forward T-S norm network in which connection weights are just the weights of weighted fuzzy if-then rules, and then trains the T-S norm neural network by a derived gradient descent algorithm. Numerical experiments show that the proposed approach is feasible and quite effective. The main contribution of this correspondence is that the mapping relationship between a set of weighted fuzzy if-then rules and a T-S norm neural network is discovered so that the difficult problem of weight acquisition in weighted fuzzy if-then rules can be converted into the training of a T-S norm neural network. A comparison between our T-S norm neural network system and a similar model (NEFCLASS) is made.