A Prototype Based Symbolic Concept Learning System
Abstract
This paper describes PROTO-TO, a cognitively based symbolic concept learning system. PROTO-TO represents concepts by learning prototypes. These prototypes are manipulated to form augmented prototypes which, when combined with a distance metric, can classify unseen instances. PROTO-TO is compared to two other concept learning systems, C4 and NTGrowth, and is shown to exhibit a performance accuracy that is comparable to C4's and superior to NTGrowth's. Unlike C4 and NTGrowth, PROTO-TO learns salient features instead of discriminant features. Furthermore, PROTO-TO does not memorize instances and then generalize its representation like C4 and NTGrowth. Rather, the mapping from instances to prototypes provides the appropriate level of generalization.
Cite
Text
de la Maza. "A Prototype Based Symbolic Concept Learning System." International Conference on Machine Learning, 1991. doi:10.1016/B978-1-55860-200-7.50012-XMarkdown
[de la Maza. "A Prototype Based Symbolic Concept Learning System." International Conference on Machine Learning, 1991.](https://mlanthology.org/icml/1991/delamaza1991icml-prototype/) doi:10.1016/B978-1-55860-200-7.50012-XBibTeX
@inproceedings{delamaza1991icml-prototype,
title = {{A Prototype Based Symbolic Concept Learning System}},
author = {de la Maza, Michael},
booktitle = {International Conference on Machine Learning},
year = {1991},
pages = {41-45},
doi = {10.1016/B978-1-55860-200-7.50012-X},
url = {https://mlanthology.org/icml/1991/delamaza1991icml-prototype/}
}