Skip to Main Content
Soft computing can be a very attractive alternative to a purely digital system, but there are many traps waiting for researchers trying to apply this new exciting technology. For nonlinear processing both neural networks and fuzzy systems can be used. Terrifically neural networks should provide much better solutions: smoother surfaces, larger number of inputs and outputs, better generalization abilities, faster processing time, etc. In industrial practice, however, many people are frustrated with neural networks not being aware that the reason for their frustrations are wrong learning algorithms and wrong neural network architectures. Having difficulties with neural network training, many industrial practitioner are enlarging neural networks and indeed such networks converges to solutions much faster. But at the same time such excessively large network are not able to respond correctly to new patterns which were not used for training. This paper describes how to use effective neural networks and how to avoid all reasons for frustration.