Synthesize Models for Quantitative Analysis Using Automata Learning.

We apply a probably approximately correct learning algorithm for multiplicity automata to generate quantitative models of system behaviors with a statistical guarantee. Using the generated model, we give two analysis algorithms to estimate the minimum and average values of system behaviors. We show how to apply the learning algorithm even when the alphabet is not fixed. The experimental result is encouraging; the estimation made by our approach is almost as precise as the exact reference answer obtained by a brute-force enumeration.