The model of Local Dfifferential Privacy (LDP) offers powerful privacy guarantees, and has been the subject of much study in the last few years, due in part to its adoption by various large technology companies. The principal applications have been in collecting frequency statistics, and finding frequent items over large domains, by combining “frequency oracles” with sketching and heavy hitter techniques. In this talk, I'll briefly recap LDP and frequent items techniques. I'll then focus on recent work on building multidimensional data models and cumulative frequency distributions. These foundational problems are helpful steps to support the emerging notion of 'Federated Analytics'. Finally, I'll mention some ongoing extensions to emerging privacy models including the shuffle model and space-bounded zero knowledge.
This file was generated by bibtex2html 1.92.