Improved Training via Incremental Learning
Abstract
It is well known that it is possible to choose training instances more carefully when using incremental learning than when using nonincremental learning. This is because a partially learned concept can provide guidance about which instances would be informative training instances. The point is illustrated by comparing the incremental decision tree induction algorithm ID5R with ID3 on Quinlan's chess task. The instance selection strategy for ID5R finds a set of training instances that produces the smallest decision tree yet found for the chess task.
Cite
Text
Utgoff. "Improved Training via Incremental Learning." International Conference on Machine Learning, 1989. doi:10.1016/B978-1-55860-036-2.50092-8Markdown
[Utgoff. "Improved Training via Incremental Learning." International Conference on Machine Learning, 1989.](https://mlanthology.org/icml/1989/utgoff1989icml-improved/) doi:10.1016/B978-1-55860-036-2.50092-8BibTeX
@inproceedings{utgoff1989icml-improved,
title = {{Improved Training via Incremental Learning}},
author = {Utgoff, Paul E.},
booktitle = {International Conference on Machine Learning},
year = {1989},
pages = {362-365},
doi = {10.1016/B978-1-55860-036-2.50092-8},
url = {https://mlanthology.org/icml/1989/utgoff1989icml-improved/}
}