ML Anthology

The Minimum Description Length Principle and Categorical Theories

J. Ross Quinlan
ICML 1994 pp. 233-241
doi:10.1016/B978-1-55860-335-6.50036-2 /icml/1994/quinlan1994icml-minimum/

Abstract

Rissanen's Minimum Description Length Principle provides a clear formalism for weighing the apparent accuracy of a theory against its complexity. When learning in categorical domains, however, one common method of applying MDL can lead to theories with poor predictive accuracy. A simple additional bias has been found to improve the usefulness of MDL in such cases. The paper presents examples using artificial data and in a real-world application.

ICML Semantic Scholar

Cite

Text

Quinlan. "The Minimum Description Length Principle and Categorical Theories." International Conference on Machine Learning, 1994. doi:10.1016/B978-1-55860-335-6.50036-2

Markdown

[Quinlan. "The Minimum Description Length Principle and Categorical Theories." International Conference on Machine Learning, 1994.](https://mlanthology.org/icml/1994/quinlan1994icml-minimum/) doi:10.1016/B978-1-55860-335-6.50036-2

BibTeX

@inproceedings{quinlan1994icml-minimum,
  title     = {{The Minimum Description Length Principle and Categorical Theories}},
  author    = {Quinlan, J. Ross},
  booktitle = {International Conference on Machine Learning},
  year      = {1994},
  pages     = {233-241},
  doi       = {10.1016/B978-1-55860-335-6.50036-2},
  url       = {https://mlanthology.org/icml/1994/quinlan1994icml-minimum/}
}