This talk is a continuation of the talk I gave at the FSA colloquium of 4 November 2021, where I showed how a system that consists of a collection of synchronously communicating automata can be learned in a compositional fashion. However, this requires a priori knowledge of the alphabet of each component, making the approach not truly black box. After giving a recap of this work, I will show how the algorithms can be extended so this assumption can be dropped. This required careful study of the properties of parallel composition operator we consider.

This is ongoing research in collaboration with Matteo Sammartino, Léo Henry and Mohammad Mousavi.