Comparing learning performance of neural networks and fuzzy systems

The learning performance of neural networks and fuzzy systems is compared. Results obtained using neural networks and fuzzy systems in three problems are presented predicting a chaotic time series, identifying a nonlinear dynamical system, and learning inverse kinematics in robot control. Simulations show that fuzzy systems can usually be trained several orders of magnitude faster than neural networks trained by the now-classical backpropagation method and that their performance equals, if not exceeds, that of neural networks