Learning Invariants from Explanations
Abstract
We present a new Explanation-Based Learning technique that learns a new kind of plan knowledge: invariant patterns in the application domain. These invariants are not plans nor macro-operators, they are not goal oriented, and they can be used by nonlinear planners. We describe the LIFE system that automatically discovers, learns and uses invariants. The paper contains a theoretical analysis of the ‘learning about invariant’ problem based on temporal generalization. One of its most distinguishing feature is the use of mathematical induction for reasoning about impossibility. This can be applied in intractable domains. The resulting learning method relies on a new kind of explanation structure, called blocking graph, which is in no way tied to a problem solving trace. This method is contrasted with standard EBL applied to planning, and some experimental results are provided.
Cite
Text
Puget. "Learning Invariants from Explanations." International Conference on Machine Learning, 1989. doi:10.1016/B978-1-55860-036-2.50057-6Markdown
[Puget. "Learning Invariants from Explanations." International Conference on Machine Learning, 1989.](https://mlanthology.org/icml/1989/puget1989icml-learning/) doi:10.1016/B978-1-55860-036-2.50057-6BibTeX
@inproceedings{puget1989icml-learning,
title = {{Learning Invariants from Explanations}},
author = {Puget, Jean-Francois},
booktitle = {International Conference on Machine Learning},
year = {1989},
pages = {200-204},
doi = {10.1016/B978-1-55860-036-2.50057-6},
url = {https://mlanthology.org/icml/1989/puget1989icml-learning/}
}