Information-Theoretic Privacy in Federated Submodel learning

TL;DR

This paper introduces an information-theoretic privacy scheme for federated submodel learning, offering stronger privacy guarantees than traditional secure aggregation, and compares its efficiency with naive approaches.

Contribution

It proposes a novel scheme combining private information retrieval with federated submodel learning to enhance privacy and efficiency.

Findings

Achieves minimal download in privacy-preserving submodel retrieval

Outperforms naive approaches in communication overhead

Provides theoretical analysis of privacy guarantees

Abstract

We consider information-theoretic privacy in federated submodel learning, where a global server has multiple submodels. Compared to the privacy considered in the conventional federated submodel learning where secure aggregation is adopted for ensuring privacy, information-theoretic privacy provides the stronger protection on submodel selection by the local machine. We propose an achievable scheme that partially adopts the conventional private information retrieval (PIR) scheme that achieves the minimum amount of download. With respect to computation and communication overhead, we compare the achievable scheme with a naive approach for federated submodel learning with information-theoretic privacy.