Efficient Implementation of Conformance Testing Techniques in Active Automata Learning

Abstract

This thesis proposes conformance testing algorithms, in the context of Active Automata Learning, that leverage knowledge from previous learning rounds to improve the efficiency of counterexample search. The focus is on evaluating the impact of targeting the newly learned states of a hypothesis. Three modifications of existing algorithms—the W Method, the Wp Method, and the Random Wp Method—were tested alongside a novel approach, the Stochastic State Coverage Method. Experimental results show that the modification of the Wp Method outperforms its original version. In contrast, the modifications of the W Method and the Random Wp Method, as well as the newly proposed method, yielded mixed results, with no clear overall improvement across all scenarios. This study shows that targeting the new states of a hypothesis may improve conformance testing performance. Future work can focus on exploring the circumstances under which better results can be achieved, as well as on the development of new algorithms that follow a similar approach.