Active automata learning algorithms face challenges when dealing with conflicting information in observation data, making them less effective in scenarios with noise or system mutations. To address this, we propose the Conflict-Aware Active Automata Learning (C3AL) framework. C3AL considers the observation tree as an integral part of the learning process, minimizing the number of tests performed on the system under learning, especially when conflicts arise. We have evaluated C3AL using a diverse range of benchmarks, encompassing over 30 realistic targets and more than 18,000 scenarios. Our evaluation results demonstrate that C3AL is a suitable framework for closed-box learning that can effectively handle noise and mutations.
Active Automata Learning with Conflict Awareness (arXiv:2308.14781v1 [cs.LG])
