Difference between revisions of "Globerson 2007 exponentiated gradient algorithms for log linear structured prediction"