Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Buffered Asynchronous Secure Aggregation for Cross-Device Federated Learning

Kun Wang, Yi-Rui Yang, Wu-Jun Li

TL;DR

BASA addresses the incompatibility between secure aggregation and asynchronous cross-device federated learning by introducing a buffered, pairwise-masked SA that requires only one round of communication per user. It leverages attribute-based encryption to securely share mask seeds within a fixed buffer, enabling asynchronous aggregation without dedicated hardware. The approach demonstrates substantial speedups and scalability advantages over traditional SA protocols, particularly under straggler conditions, while preserving privacy under the honest-but-curious model. This work offers a practical pathway to privacy-preserving, scalable AFL on heterogeneous edge devices.

Abstract

Asynchronous federated learning (AFL) is an effective method to address the challenge of device heterogeneity in cross-device federated learning. However, AFL is usually incompatible with existing secure aggregation protocols used to protect user privacy in federated learning because most existing secure aggregation protocols are based on synchronous aggregation. To address this problem, we propose a novel secure aggregation protocol named buffered asynchronous secure aggregation (BASA) in this paper. Compared with existing protocols, BASA is fully compatible with AFL and provides secure aggregation under the condition that each user only needs one round of communication with the server without relying on any synchronous interaction among users. Based on BASA, we propose the first AFL method which achieves secure aggregation without extra requirements on hardware. We empirically demonstrate that BASA outperforms existing secure aggregation protocols for cross-device federated learning in terms of training efficiency and scalability.

Buffered Asynchronous Secure Aggregation for Cross-Device Federated Learning

TL;DR

BASA addresses the incompatibility between secure aggregation and asynchronous cross-device federated learning by introducing a buffered, pairwise-masked SA that requires only one round of communication per user. It leverages attribute-based encryption to securely share mask seeds within a fixed buffer, enabling asynchronous aggregation without dedicated hardware. The approach demonstrates substantial speedups and scalability advantages over traditional SA protocols, particularly under straggler conditions, while preserving privacy under the honest-but-curious model. This work offers a practical pathway to privacy-preserving, scalable AFL on heterogeneous edge devices.

Abstract

Asynchronous federated learning (AFL) is an effective method to address the challenge of device heterogeneity in cross-device federated learning. However, AFL is usually incompatible with existing secure aggregation protocols used to protect user privacy in federated learning because most existing secure aggregation protocols are based on synchronous aggregation. To address this problem, we propose a novel secure aggregation protocol named buffered asynchronous secure aggregation (BASA) in this paper. Compared with existing protocols, BASA is fully compatible with AFL and provides secure aggregation under the condition that each user only needs one round of communication with the server without relying on any synchronous interaction among users. Based on BASA, we propose the first AFL method which achieves secure aggregation without extra requirements on hardware. We empirically demonstrate that BASA outperforms existing secure aggregation protocols for cross-device federated learning in terms of training efficiency and scalability.
Paper Structure (30 sections, 1 theorem, 11 equations, 4 figures, 1 table, 3 algorithms)

This paper contains 30 sections, 1 theorem, 11 equations, 4 figures, 1 table, 3 algorithms.

Table of Contents

  1. Introduction
  2. Background
  3. Federated Learning
  4. Asynchronous Federated Learning
  5. Secure Aggregation
  6. Attribute-based Encryption
  7. Method
  8. Intuition
  9. Pairwise Mask-based Framework
  10. ABE-based Random Seed Sharing
  11. Detailed Protocol
  12. AFL with BASA
  13. Security Analysis
  14. Threat Model
  15. Security Guarantee
  16. ...and 15 more sections

Key Result

Theorem 3.1

In BASA, the server can only learn the partial aggregation result over the inputs of honest users $\sum_{i \in \mathcal{U} \setminus \mathcal{S}} \mathbf{x}_i$, under the honest-but-curious threat model.

Figures (4)

  • Figure 1: A brief illustration of the pairwise mask-based framework of BASA with a simple case study of three users. Each user uploads local inputs in turn from left to right. $\mathbf{y}_1, \mathbf{y}_2, \mathbf{y}_3$ are the masked inputs, and $s_{12}, s_{13}, s_{23}$ are encrypted ciphertexts of the corresponding random seeds.
  • Figure 2: The wall-clock time (in second) to reach a target validation accuracy for 200 users when training CNN models on the FEMNIST and CIFAR-10 datasets with different delay scales.
  • Figure 3: The wall-clock time (in second) taken by a user to execute the BASA protocol and LSA protocol with different user numbers and buffer sizes. The terms "$0.1$million", "$0.5$million", and "$1.0$million" refer to different model dimensions.
  • Figure 4: The wall-clock time (in second) required to reach a target validation accuracy for training CNN models on the FEMNIST dataset with different numbers of concurrent training users.

Theorems & Definitions (2)

  • Theorem 3.1
  • proof