comments and criticism

We would like to thank the reviewers for their comments and helpful suggestions. We respond below to the reviewers' R1: The term "privacy model" is confusing: We will definitely think of a better term. R1: More discussion of the broader impact" We will elaborate more on the potential impact of the proposed We believe it can be a basis for a more flexible, more expressive framework for DP learning. R3: "Intuitively, the sample complexity should be comparable to that of the non-private one": We disagree with the R3: "I'd expect some simple but a bit more advance model, like a Bernoulli model . . . We also want to point out that the label-determined model does capture some realistic scenarios.