Scalable Private Learning with PATE

   Abstract