Title :
Definition of initial tuning parameters by using fuzzy-exceeding ball clustering method
Author :
Liu, Wen-yuan ; Ma, Kun ; Deng, Cheng-yu ; Wang, Bao-wen ; Shi, Yan ; Fang, Shu-fen
Author_Institution :
Sch. of Manage., Harbin Inst. of Technol., China
Abstract :
Very few of the suitable initial values of tuning parameters are argued in neuro-fuzzy algorithms, which are often used, so this can affect the nicety of the neuro-fuzzy algorithm. Although we can design initial tuning parameters by using the fuzzy c-means clustering algorithm before learning the corresponding fuzzy rules, the number of pattern collection must be known firstly. Thereby, we band the idea of fuzzy-exceeding ball with neuro-fuzzy network together, and adjust number, centers and widths of the ball, optimize the border pattern collection to confirm the weight values of parameters. We can minimize error and improve nicety of algorithm by using it.
Keywords :
fuzzy neural nets; optimisation; pattern clustering; border pattern collection; fuzzy c-means clustering; fuzzy rules; fuzzy-exceeding ball clustering method; initial tuning parameters; neuro-fuzzy algorithms; Clustering algorithms; Clustering methods; Engineering management; Fuzzy neural networks; Management information systems; Neural networks; Noise generators; Technology management; Training data; Transfer functions;
Conference_Titel :
Machine Learning and Cybernetics, 2003 International Conference on
Print_ISBN :
0-7803-8131-9
DOI :
10.1109/ICMLC.2003.1259640