Domain Compression: A primitive for distributed inference under communication & privacy constraints

A Google TechTalk, 2020/7/30, presented by Jayadev Acharya, Cornell University ABSTRACT: We present Domain Compression, a method for reducing the alphabet size of statistical problems. It is randomness efficient, which makes it particularly suitable in distributed settings with constraints such as privacy and communication. We illustrate the method with a distributed goodness-of-fit algorithm that is sample-, communication-,privacy-, and public randomness- optimal, all simultaneously.