A learning-based algorithm selection meta-reasoner for the real-time MPE problem

The algorithm selection problem aims to select the best algorithm for an input problem instance according to some characteristics of the instance. This paper presents a learning-based inductive approach to build a predictive algorithm selection system from empirical algorithm performance data of the Most Probable Explanation(MPE) problem. The learned model can serve as an algorithm selection meta-reasoner for the real-time MPE problem. Experimental results show that the learned algorithm selection models can help integrate multiple MPE algorithms to gain a better overall performance of reasoning