An approach to rule-based knowledge extraction

The extraction of easily interpretable knowledge from the large amount of data measured in experiments is well desirable. This paper proposes a method to achieve this. A fuzzy rule system is first generated and optimized using evolution strategies. This fuzzy system is then converted to an RBF neural network to refine the obtained knowledge. In order to extract understandable fuzzy rules from the trained RBF network, a neural network regularization technique called adaptive weight sharing is developed. Simulation results on the Mackey-Glass system show that the proposed approach to knowledge extraction is effective and practical.