CrySSMEx, is an algorithm for extracting Finite State Machines from Recurrent Neural Networks. Input: sequential data generated from RNNs. Output: (stochastic) FSMs and state space quantizers.
This homepage is still under construction. More information on sourceforge.net/projects/cryssmex. Version 0.1 of the source code is now released.
An article on the algorithm is accepted for publication: [pdf]. Another article, "Rule Extraction from Recurrent Neural Networks: A Taxonomy and Review" provides some background to the field [pdf].
Project manager: Henrik Jacobsson