![[2602.21674] Error-awareness Accelerates Active Automata Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21674] Error-awareness Accelerates Active Automata Learning
Summary
The paper discusses how error-awareness can enhance Active Automata Learning (AAL) algorithms, enabling them to learn more efficiently from systems with observable errors, thus improving scalability.
Why It Matters
This research addresses a critical challenge in machine learning related to scaling AAL algorithms. By leveraging error-awareness, the findings could significantly improve the efficiency of learning models in various applications, making it relevant for researchers and practitioners in the field of machine learning.
Key Takeaways
- Error-awareness can significantly accelerate Active Automata Learning.
- The study provides adaptations of the state-of-the-art AAL algorithm L# based on varying degrees of domain knowledge.
- Empirical evaluations show substantial improvements in learning efficiency with realistic domain knowledge.
Computer Science > Machine Learning arXiv:2602.21674 (cs) [Submitted on 25 Feb 2026] Title:Error-awareness Accelerates Active Automata Learning Authors:Loes Kruger, Sebastian Junges, Jurriaan Rot View a PDF of the paper titled Error-awareness Accelerates Active Automata Learning, by Loes Kruger and 2 other authors View PDF HTML (experimental) Abstract:Active automata learning (AAL) algorithms can learn a behavioral model of a system from interacting with it. The primary challenge remains scaling to larger models, in particular in the presence of many possible inputs to the system. Modern AAL algorithms fail to scale even if, in every state, most inputs lead to errors. In various challenging problems from the literature, these errors are observable, i.e., they emit a known error output. Motivated by these problems, we study learning these systems more efficiently. Further, we consider various degrees of knowledge about which inputs are non-error producing at which state. For each level of knowledge, we provide a matching adaptation of the state-of-the-art AAL algorithm L# to make the most of this domain knowledge. Our empirical evaluation demonstrates that the methods accelerate learning by orders of magnitude with strong but realistic domain knowledge to a single order of magnitude with limited domain knowledge. Subjects: Machine Learning (cs.LG); Logic in Computer Science (cs.LO) Cite as: arXiv:2602.21674 [cs.LG] (or arXiv:2602.21674v1 [cs.LG] for this version) https:...
