Large-scale Learning With Global Non-Decomposable Objectives

   Abstract