Rules of chaotic behaviour extracted from a fuzzy-neural network

The paper presents a fuzzy-neuro architecture called FuNN (fuzzy-neural network) along with a structural learning algorithm based on backpropagation with forgetting. The FuNN structure is adaptable and it can automatically capture the main characteristics and the underlying rules of the problem. Adaptation includes both the membership functions and rules of the fuzzy system. The key point of the paper is to show that the combined use of the fuzzy rule-based system with structural learning results in a powerful and efficient tool of automatic extraction of meaningful rules. One important application of the methodology introduced is neuro-embedding, which is a novel way of describing chaotic processes. Neuro-embedding is illustrated on the example of modelling and prediction of the Mackey-Glass data set.