Artificial Universes - Towards a Systematic Approach to Evaluation Algorithms Which Learn Form Examples
Abstract
The theory behind artificial universes - full probabilistic models of domains - is developed. Universes can be used to provide a ready supply of randomly selected examples for use by classification algorithms. Simple universes are quite sufficient to provide challenging tasks for such algorithms. The means of defining a universe is described in detail with a running example. The issue of alternate representation is discussed and the idea of a standard representation such as the ID3 form is introduced. Properties of attributes, especially those of pure noise and redundancy, are defined. The use of information theory concepts (including the primitive notion of majorisation) to model noise in the domain and to provide measures for the universe is described. In particular the notions of accuracy and efficiency of an induced rule are defined.
Cite
Text
Hickey. "Artificial Universes - Towards a Systematic Approach to Evaluation Algorithms Which Learn Form Examples." International Conference on Machine Learning, 1992. doi:10.1016/B978-1-55860-247-2.50030-9Markdown
[Hickey. "Artificial Universes - Towards a Systematic Approach to Evaluation Algorithms Which Learn Form Examples." International Conference on Machine Learning, 1992.](https://mlanthology.org/icml/1992/hickey1992icml-artificial/) doi:10.1016/B978-1-55860-247-2.50030-9BibTeX
@inproceedings{hickey1992icml-artificial,
title = {{Artificial Universes - Towards a Systematic Approach to Evaluation Algorithms Which Learn Form Examples}},
author = {Hickey, Ray J.},
booktitle = {International Conference on Machine Learning},
year = {1992},
pages = {196-205},
doi = {10.1016/B978-1-55860-247-2.50030-9},
url = {https://mlanthology.org/icml/1992/hickey1992icml-artificial/}
}