| Autor: |
Wang, Jun, Liao, Xiaofeng, Yi, Zhang, Chen, Yong, Wang, Guoyin, Jin, Fan, Yan, Tianyun |
| Zdroj: |
Advances in Neural Networks - ISNN 2005 (9783540259121); 2005, p455-460, 6p |
| Abstrakt: |
All existing architectures and learning algorithms for Generalized Congruence Neural Network (GCNN) seem to have some shortages or lack rigorous theoretical foundation. In this paper, a novel GCNN architecture (BPGCNN) is proposed. A new error back-propagation learning algorithm is also developed for the BPGCNN. Experimental results on some benchmark problems show that the proposed BPGCNN performs better than standard sigmoidal BPNN and some improved versions of BPNN in convergence speed and learning capability, and can overcome the drawbacks of other existing GCNNs. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Supplemental Index |
| Externí odkaz: |
|
