Parallel distributed processing: explorations in the microstructure of cognition, vol. 2: psychological and biological models
Introduction to the Theory of Neural Computation
Introduction to the Theory of Neural Computation
Efficient classification for multiclass problems using modular neural networks
IEEE Transactions on Neural Networks
An accelerated learning algorithm for multilayer perceptrons: optimization layer by layer
IEEE Transactions on Neural Networks
MICAI'11 Proceedings of the 10th international conference on Artificial Intelligence: advances in Soft Computing - Volume Part II
Hi-index | 0.00 |
Initial learning process of the BP, which can influence the performance oflearning in multiclass classification problems, is analyzed. Also, theweights decreasing phenomena in the initial stage of learning areinvestigated. On the basis of this analysis, a new initialization methodis proposed. The proposed method minimizes the initial objective function.It eliminates the phenomenon that weights decrease in the beginning oflearning. Several simulation results show that the proposed initializationmethod performs much better than the conventional random initializationmethod in the batch mode and slightly better in the pattern mode. Since itrequires only a little additional computation, it is a strong alternativeto the conventional random initialization. It is expected that theproposed initialization method can be used with any accelerated learningalgorithm to enhance the learning speed.