Abductive explanation-based learning: A solution to the multiple inconsistent explanation problem (Q1189662)

Frequently, as an effect of the incompleteness and incorrectness of the available information, domain theories produce multiple explanations for training instances, not all of them are correct. The issue proposes the abductive explanation-based learning (A-EBL) as an attempt to solve the multiple inconsistent explanation problem mainly based on using negative examples and set covering techniques. The EBL/TS and the abductive-EBL algorithms are presented in the second section. A formal treatment of A-EBL and EBL/TS, expressed in terms of theoretical results concerning competence and convergence properties, is developed in the next section. Some experimental results on A-EBL are also reported. It is proved that the A-EBL has convergence properties only logarithmically worse than EBL/TS and that A-EBL is computationally efficient when the domain theory is tractable.