Robust by Design: SAT-based Automata Learning in the Presence of Noise

Description

Automata learning has long since excelled in learning behaviour models from black-box systems. For this, a lot of different methods exists, among them, SAT solving can usually be used to exactly infer deterministic automata from a set of execution traces. However, in practice, systems and data sets may not be perfectly deterministic and may often contain faults due to message loss or other environmental factors. We present a method, using partial Max-SAT, to learn deterministic models from noisy execution traces and present current research to apply this to active automata learning as well.

Period	26 Sept 2025
Event title	AVM25: 17th Alpine Verification Meeting
Event type	Conference
Location	Timisoara, RomaniaShow on map
Degree of Recognition	International