Explanation-Based Generalization: A Unifying View
Machine Learning
| Hi-index | 0.00 |
LAIR is a system that incrementally learns conjunctive concept descriptions from positive and negative examples. These concept descriptions are used to create and extend a domain theory that is applied, by means of constructive induction, to later learning tasks. Important issues for constructive induction are when to do it and how to control it LAIR demonstrates how constructive induction can be controlled by (1) reducing it to simpler operations, (2) constraining the simpler operations to preserve relative correctness, (3) doing deductive inference on an as-needed basis to meet specific information requirements of learning subtasks, and (4) constraining the search space by subtask-dependent constraints.