Abstract
Multilayer perceptrons have been shown to approximate any continuous functions with a desired precision. With insufficient training samples, however, the network can not learn the function properly and popular model selection methods such as cross validation can not be used. We propose a scheme to generate virtual samples using a population of networks. They are applied to regression problems and are shown to improve generalization and to solve the model selection problem at the same time.
Similar content being viewed by others
References
G. An, “The effects of adding noise duringbackpropagation training on a generalization performance”, Neural Computation, Vol. 7, pp. 613–674, 1996.
A. Bandura, SocialLearning Theory, General Learning Press: New York, 1971.
P. Bartlett, “Vapnik-Chervonenkis dimension bounds for two-and three-layer networks”, Neural Computation, Vol. 5, pp. 371–373, 1993.
S. Cho and K. Cha, “Evolution of neural network training setthrough addition of virtual samples”, Proceedings 1996 IEEE International Conference on Evolutionary Computation, pp. 685–688, Nagoya, Japan, 1996.
D. Cohn, Z. Ghahramani and M. Jordan, “Active learning withstatistical models”, Neural Information Processing Systems, Vol. 7, pp. 705–712, 1995.
R. Duda and P. Hart, Pattern Classificationand Scene Analysis, Wiley: New York, 1973.
K. Funahashi,“On the approximate realization of continuous mappings by neural networks”, Neural Networks, Vol. 2, pp. 189–195, 1989.
A. Abu-Mostafa, “Hints”, Neural Computation, Vol.7, pp. 639–671, 1995.
M. Perrone, “Generalaveraging results for convex optimization”, in M. Mozer et al. (eds), Proceedings 1993 Connectionist Models Summer School, pp. 364–371, Lawrence Erlbaum: Hillsdale, USA, 1994.
Author information
Authors and Affiliations
Rights and permissions
About this article
Cite this article
Cho, S., Jang, M. & Chang, S. Virtual sample generation using a population of networks. Neural Processing Letters 5, 21–27 (1997). https://doi.org/10.1023/A:1009653706403
Issue Date:
DOI: https://doi.org/10.1023/A:1009653706403