IEEE-INNS-ENNS International Joint Conference on Neural Networks (IJCNN'00)-Volume 3
Weight Groupings in the Training of Recurrent Networks
Como, Italy
July 24-July 27
ISBN: 0-7695-0619-4
In this paper, we use the block-diagonal matrix to approximate the Hessian matrix in the Levenberg Marquardt method in the training of recurrent neural networks. Substantial improvement of the training time over the original Levenberg Marquardt method is observed without degrading the generalization ability. Three weight grouping methods, correlation blocks, k-unit blocks and layer blocks were investigated and compared. Their computational complexity, approximation ability, and training time are analyzed.
Citation:
Lai-Wan Chan, Chi-Cheong Szeto, "Weight Groupings in the Training of Recurrent Networks," ijcnn, vol. 3, pp.3021, IEEE-INNS-ENNS International Joint Conference on Neural Networks (IJCNN'00)-Volume 3, 2000