Residuality and Learning for Register Automata

18 Sep 2020
Joshua Moerman
Highlights 2020


In this research we consider the problem of inferring a register automaton from observations. This has been done before for deterministic RA, but is still open for nondeterministic RA. To see why nondeterminism is interesting, consider the well-known learning algorithms L* and NL* for respectively deterministic and nondeterministic automata. Although the representation is different, they operate on the same class of languages (i.e., regular languages). This is not the case for RA, where nondeterminism gives a strictly bigger class of languages than determinism. So not only does the representation changes, so does the class of languages. Our contributions are as follows. This is joint work with Matteo Sammartino.

Extended abstract (PDF)
Poster (PDF)
Full paper