Discovering interpretable classification rules from neural processed data