On Resilient and Efficient Linear Secure Aggregation in Hierarchical Federated Learning

TL;DR

This paper investigates the fundamental communication and randomness costs for secure aggregation in hierarchical federated learning with unreliable links, proposing an optimal protocol and bridging theory with practical applications.

Contribution

It characterizes the minimum costs for secure aggregation under unreliable communication and introduces an optimal protocol with a new problem formulation connecting theory and practice.

Findings

Minimum communication and randomness costs are established.

An optimal secure aggregation protocol is proposed and proven to be optimal.

A new problem formulation bridges theoretical and practical federated learning scenarios.

Abstract

In this paper, we study the fundamental limits of hierarchical secure aggregation under unreliable communication. We consider a hierarchical network where each client connects to multiple relays, and both client-to-relay and relay-to-server links are intermittent. Under this setting, we characterize the minimum communication and randomness costs required to achieve robust secure aggregation. We then propose an optimal protocol that attains these minimum costs, and establish its optimality through a matching converse proof. In addition, we introduce an improved problem formulation that bridges the gap between existing information-theoretic secure aggregation protocols and practical real-world federated learning problems.