NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:7289
Title:Privacy Amplification by Mixing and Diffusion Mechanisms

The work gives a collection of results on privacy amplification by postprocessing. The results reveal some interesting connections. Further, the work gives a refinement of the analysis from Feldman et al, 2018 work on privacy amplification by iteration in the strongly convex case. While the initial reviews were very strong, an additional solicited review by an expert has brought up significant issues. Specifically, misrepresentation of the strongly convex case as a new application, lack of any other new applications and poor presentation in the section on diffusion. While the eventual decision is to accept the work these issues should be addressed as well as possible in the final version.