An ε-insensitive Learning in Neuro-Fuzzy Modeling
Jacek M. Łęski, Norbert Henzel · 2003
The neuro-fuzzy modeling has an intrinsic inconsistency. It may perform thinking tolerant to imprecision, but neural networks learning methods are zero-tolerant to imprecision. Proposed method make it possible to exclude this intrinsic inconsistency of neuro-fuzzy modeling. This new method can be called ε -insensitive learning or ε learning, where in order to fit fuzzy model to real data, ε -insensitive loss function is used. Computationally efficient numerical method for the ε -insensitive learning is proposed. Finally, numerical example is given to demonstrate the improved generalization ability of obtained fuzzy model.