Is learning RFSAs better than learning DFAs?

Abstract

Inference of RFSAs has been recently presented [1] as an alternative to inference of DFAs if the target language has been obtained by a random generation of NFAs. We propose in this paper the algorithm RPNI2, which is a variation of the previous RPNI, that also outputs DFAs as hypothesis. The experiments done using the same data as in [1] show that RPNI2 has an error rate very similar to the rate obtained in the inference of RFSAs, but the size of the hypothesis is substantially smaller. © Springer-Verlag Berlin Heidelberg 2006.