Input data clustering to improve neural network performance

We focus on the pre-processing of the training data. We find that when a network is trained with selected input, the performance of the network improves significantly as opposed to a network that does not receive selected input data for training. Furthermore, less time is required to train such networks. The problem of image deblurring is used to test the performance of such a network.

[1]  Min Su,et al.  A hybrid learning system for image deblurring , 2002, Pattern Recognit..

[2]  Daniel N. Osherson,et al.  Modular learning , 1993 .

[3]  J. R. Hughes.  Cerebral lateralization: biological mechanisms, associations and pathology , 1987 .

[4]  Geoffrey E. Hinton,et al.  Adaptive Mixtures of Local Experts , 1991, Neural Computation.

[5]  M. Basu,et al.  Gating improves neural network performance , 2001, IJCNN'01. International Joint Conference on Neural Networks. Proceedings (Cat. No.01CH37222).

[6]  S. Kosslyn.  Seeing and imagining in the cerebral hemispheres: a computational approach. , 1987, Psychological review.

[7]  Y. V. Venkatesh,et al.  Deblurring Gaussian blur using a wavelet array transform , 1995, Pattern Recognit..

[8]  Michael I. Jordan,et al.  Convergence results for the EM approach to mixtures of experts architectures , 1995, Neural Networks.