Scalable Private Learning with PATE

  Abstract