Some Interesting Properties of a Connectionist Inductive Learning System
Abstract
We present a connectionist inductive system that displays the following behavior: i) as information in test examples increases, the system's categorization time decreases, ii) with amount of information held constant, the system categorizes examples with partially–specified feature values faster than examples with completely-specified feature values, iii) the system detects novel features of objects that deviate from its expectations. Our system's performance on test examples can be evaluated in terms of both categorization accuracy and certainty. We discuss these properties (and previously demonstrated properties of the system) in terms of their psychological plausibility and their desirability for a system functioning in real-world domains.
Cite
Text
Wisniewski and Anderson. "Some Interesting Properties of a Connectionist Inductive Learning System." International Conference on Machine Learning, 1988. doi:10.1016/B978-0-934613-64-4.50024-4Markdown
[Wisniewski and Anderson. "Some Interesting Properties of a Connectionist Inductive Learning System." International Conference on Machine Learning, 1988.](https://mlanthology.org/icml/1988/wisniewski1988icml-some/) doi:10.1016/B978-0-934613-64-4.50024-4BibTeX
@inproceedings{wisniewski1988icml-some,
title = {{Some Interesting Properties of a Connectionist Inductive Learning System}},
author = {Wisniewski, Edward J. and Anderson, James A.},
booktitle = {International Conference on Machine Learning},
year = {1988},
pages = {181-187},
doi = {10.1016/B978-0-934613-64-4.50024-4},
url = {https://mlanthology.org/icml/1988/wisniewski1988icml-some/}
}