Relations, Knowledge and Empirical Learning
Abstract
Recent studies in empirical learning have sought to use domain knowledge in addition to training examples. We generalize these strategies for finding the deductive biases arising from domain knowledge. We introduce a relational framework, in which domain knowledge, concepts and newly constructed features are all related directly by viewing them as relations. Knowledge is used to identify high-level similarities (relations) in the training examples, to construct new features that are more suitable for learning the concept. Results on the utility of knowledge and feature construction with the DRL algorithm are discussed.
Cite
Text
Ragavan and Rendell. "Relations, Knowledge and Empirical Learning." International Conference on Machine Learning, 1991. doi:10.1016/B978-1-55860-200-7.50041-6Markdown
[Ragavan and Rendell. "Relations, Knowledge and Empirical Learning." International Conference on Machine Learning, 1991.](https://mlanthology.org/icml/1991/ragavan1991icml-relations/) doi:10.1016/B978-1-55860-200-7.50041-6BibTeX
@inproceedings{ragavan1991icml-relations,
title = {{Relations, Knowledge and Empirical Learning}},
author = {Ragavan, Harish and Rendell, Larry A.},
booktitle = {International Conference on Machine Learning},
year = {1991},
pages = {188-192},
doi = {10.1016/B978-1-55860-200-7.50041-6},
url = {https://mlanthology.org/icml/1991/ragavan1991icml-relations/}
}