Improving Register Automata Learning by Leveraging Testing Methods for Finite State Machines

Active automata learning is a technique for modeling the runtime behaviour of software and hardware components. Finite state machines have long been used to model control flow. Register automata can also describe data flow but active automata learning with register automata does not scale well. This project aims to optimise register automata learning by leveraging testing algorithms for finite state machines. A hypothesised register automaton is abstracted to a finite state machine to generate counterexamples. These are translated for the register automaton to make the structure of the system emerge. After the learning converges, a standard method is used to further refine the register automaton model. Performance of the new algorithm is evaluated against the current best available option. Results show that the new oracle finds a sufficient number of locations to be of theoretical interest in register automata testing, but in practice the number of resets and inputs are uncertain, mainly due to restrictions of the method used in the refinement phase.