Abstract
A new training approach based on the Levenberg-Marquardt algorithm is proposed for type-2 fuzzy neural networks. While conventional gradient descent algorithms use only the first order derivative, the proposed algorithm used in this paper benefits from the first and the second order derivatives which makes the training procedure faster. Besides, this approach is more robust than the other techniques that use the second order derivatives, e.g. Gauss-Newton's method. The training algorithm proposed is tested on the training of a type-2 fuzzy neural network used for the prediction of a chaotic Mackey-Glass time series. The results show that the learning algorithm proposed not only results in faster training but also in a better forecasting accuracy.
Original language | English |
---|---|
Title of host publication | IEEE SSCI 2011 : Symposium Series on Computational Intelligence - T2FUZZ 2011: 2011 IEEE Symposium on Advances in Type-2 Fuzzy Logic Systems |
Number of pages | 6 |
Publication date | 2011 |
Pages | 88-93 |
Article number | 5949558 |
ISBN (Print) | 9781612840789 |
DOIs | |
Publication status | Published - 2011 |
Externally published | Yes |
Event | 2011 IEEE Symposium Series on Computational Intelligence - Paris, France Duration: 11 Apr 2011 → 15 Apr 2011 |
Conference
Conference | 2011 IEEE Symposium Series on Computational Intelligence |
---|---|
Country/Territory | France |
City | Paris |
Period | 11/04/2011 → 15/04/2011 |
Sponsor | IEEE |
Keywords
- Levenberg-Marquardt algorithm
- Mackey-Glass time series
- Type-2 fuzzy neural networks