Incorporating redundant learned rules: a preliminary formal analysis of EBL
Russell Greiner, Joseph Likuski · International Joint Conference on Artificial Intelligence · 1989
Explanation-based learning -- i.e., incorporating new redundant rules suggested by earlier problem solving experiences -- is an attempt to speed up problem solving. Unfortunately, the resulting systems are not always more efficient on subsequent problems. This paper describes, analytically, whether these new rules should be added, and if so, where they should appear in the overall derivation strategy. While this task is intractable in general, we present several interesting special cases which can be solved in time (essentially) linear in the number of rules in the system.