Jurriaan Rot

Florian Frank, Stefan Milius, Jurriaan Rot et al.

Automata over infinite alphabets have emerged as a convenient computational model for processing structures involving data, such as nonces in cryptographic protocols or data values in XML documents. We introduce active learning methods for bar automata, a species of automata that process finite data words represented as bar strings, which are words with explicit name binding letters. Bar automata have pleasant algorithmic properties. We develop a framework in which every learning algorithm for standard deterministic or nondeterministic finite automata over finite alphabets can be used to learn bar automata, with a query complexity determined by that of the chosen learner. The technical key to our approach is the algorithmic handling of $α$-equivalence of bar strings, which allows bridging the gap between finite and infinite alphabets. The principles underlying our framework are generic and also apply to bar Büchi automata and bar tree automata, leading to the first active learning methods for data languages of infinite words and finite trees.

Loes Kruger, Sebastian Junges, Jurriaan Rot

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.

Loes Kruger, Sebastian Junges, Jurriaan Rot

Active automata learning (AAL) is a method to infer state machines by interacting with black-box systems. Adaptive AAL aims to reduce the sample complexity of AAL by incorporating domain specific knowledge in the form of (similar) reference models. Such reference models appear naturally when learning multiple versions or variants of a software system. In this paper, we present state matching, which allows flexible use of the structure of these reference models by the learner. State matching is the main ingredient of adaptive L#, a novel framework for adaptive learning, built on top of L#. Our empirical evaluation shows that adaptive L# improves the state of the art by up to two orders of magnitude.