On the Effect of Instance Representation on Generalization
Abstract
This paper explains why changing the instance representation changes the ability of a class of learning algorithms to generalize from examples. The explanation is based on the correspondence between data compression and generalization. The randomness of a function with respect to a class of hypotheses is introduced as a measure of the interaction between a learning algorithm and an instance representation. Different methods of reducing randomness result in different types of representation change. Two types of representation change are identified and illustrated with examples.
Cite
Text
Saxena. "On the Effect of Instance Representation on Generalization." International Conference on Machine Learning, 1991. doi:10.1016/B978-1-55860-200-7.50043-XMarkdown
[Saxena. "On the Effect of Instance Representation on Generalization." International Conference on Machine Learning, 1991.](https://mlanthology.org/icml/1991/saxena1991icml-effect/) doi:10.1016/B978-1-55860-200-7.50043-XBibTeX
@inproceedings{saxena1991icml-effect,
title = {{On the Effect of Instance Representation on Generalization}},
author = {Saxena, Sharad},
booktitle = {International Conference on Machine Learning},
year = {1991},
pages = {198-202},
doi = {10.1016/B978-1-55860-200-7.50043-X},
url = {https://mlanthology.org/icml/1991/saxena1991icml-effect/}
}