State merging inference of finite state classifiers

Abstract:

We introduce the C-regular inference problem, consisting in inducing a set of C regular languages from samples of each language. We have chosen a finite state classifier representation of the set of regular languages, that allows to study easily the mutual exclusion of the languages.  The search space for state merging methods is studied. A constraint characterization of the set of deterministic unambiguous classifiers is given and used to design a new state merging algorithm considering not only possible mergings but also impossible ones.
 


Submitted to publication...