Learning in ERE
ERE supports multiple learning mechanisms:
- Causal theory can be refined by detecting and recovering from failures, specifically when a discrepency between what was projected to occur and what actually occured during reaction.
- Explanation-based learning has been used to acquire general preconditions to be added to an operator schema, by means of generalization.
- Search control rules can be acquired and refined through caching goal-satisfying behaviors created by the Projector. Furthermore, SCRs with varying degrees of generality can be created using a goal regression algorithm. The effect is similar to Soar's mechanism of chunking.
- Problem reduction rules can be refined when there is inappropriate problem-solving behavior, in order to reduce the probability of backtracking when a partial solution is inadequate.
These various methods serve to combine tradition analytic and inductive methods of learning; furthermore, the type of learning warranted depends upon the type of knowledge being refined and on the availability and reliability of knowledge needed to support the refinement. Furthermore, the learning mechanism operates reflexively.
To return, press HOME.
To go to the next document, press NEXT.