A Probabilistic Learning Approach for Counterexample Guided Abstraction Refinement.

The paper presents a novel probabilistic learning approach to state separation problem which occurs in the counterexample guided abstraction refinement. The method is based on the sample learning technique, evolutionary algorithm and effective probabilistic heuristics. Compared with the previous work by the sampling decision tree learning solver, the proposed method outperforms 2 to 4 orders of magnitude faster and the size of the separation set is 76% smaller on average. [ABSTRACT FROM AUTHOR]