Approximately Learning Quantum Automata.

In this paper, we provide two methods for learning measure-once one-way quantum finite automata using a combination of active learning and non-linear optimization. First, we learn the number of states of a measure-once one-way quantum automaton using a heuristic binary tree representing the different variations of a Hankel matrix. Then we use two optimization methods to learn the unitary matrices representing the transitions of the automaton. When comparing the original automaton with the one learned, we provide a new way to compute the distance on the base of the language of the combined quantum automata. Finally, we show, using experiments on a set of randomly generated quantum automata, which method performs better.