Like a dawn light scattering into the cloud sky of AI, neural network and fuzzy logic become state-of-the-art technologies in exploring the intellect. To make a judgement between both technologies, we propose an evaluation on them from the view point of learning classification. Since there are a variety of models proposed within both technologies, we focus on the most significant model, i.e., Back Propagation Network (BPN) (J. McClelland et al., 1986) and Wang's fuzzy rule generator (L.X. Wang and J.M Mendel, 1992). First in the evaluation, we introduce a gravity effect field to illustrate these two models' influence under the existence of one instance. After that, we virtually construct two classification problems and discuss the behaviors of both methods through the gravity effect field. Finally, we propose another two real examples to demonstrate the results. We conclude that Wang's method is more suitable for piecewise region classification and needs more representative or complete training samples than BPN. BPN is more training data tolerant and less network parameter sensible than that of Wang's fuzzy rule generator. However, basic instinct problems still exist, BPN behavior is more black box than fuzzy rule generator.
Fuzzy Systems Symposium, 1996. Soft Computing in Intelligent Systems and Information Processing., Proceedings of the 1996 Asian, pp.79-84