Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One-Shot Secure Aggregation: A Hybrid Cryptographic Protocol for Private Federated Learning in IoT

Imraul Emmaka, Tran Viet Xuan Phuong

TL;DR

<3-5 sentence high-level summary> The paper tackles the prohibitive communication costs of secure aggregation in IoT-enabled federated learning by introducing Hyb-Agg, a one-shot, non-interactive protocol that combines Multi-Key CKKS with ECDH-based additive masking. It proves confidentiality, integrity, and collusion resistance under RLWE and CDH assumptions, modeling the PRF via ChaCha20, and achieves constant per-client communication regardless of participant count with about a 12x expansion. Empirical evaluations on both a high-end machine and a Raspberry Pi 4 demonstrate sub-second end-to-end times on edge hardware and practical scalability for IoT deployments, even under potential server collusion with up to N-2 clients. The work highlights a viable path to scalable, privacy-preserving FL in bandwidth- and energy-constrained environments, while outlining future enhancements for dropout resilience and dynamic participation.

Abstract

Federated Learning (FL) offers a promising approach to collaboratively train machine learning models without centralizing raw data, yet its scalability is often throttled by excessive communication overhead. This challenge is magnified in Internet of Things (IoT) environments, where devices face stringent bandwidth, latency, and energy constraints. Conventional secure aggregation protocols, while essential for protecting model updates, frequently require multiple interaction rounds, large payload sizes, and per-client costs rendering them impractical for many edge deployments. In this work, we present Hyb-Agg, a lightweight and communication-efficient secure aggregation protocol that integrates Multi-Key CKKS (MK-CKKS) homomorphic encryption with Elliptic Curve Diffie-Hellman (ECDH)-based additive masking. Hyb-Agg reduces the secure aggregation process to a single, non-interactive client-to-server transmission per round, ensuring that per-client communication remains constant regardless of the number of participants. This design eliminates partial decryption exchanges, preserves strong privacy under the RLWE, CDH, and random oracle assumptions, and maintains robustness against collusion by the server and up to $N-2$ clients. We implement and evaluate Hyb-Agg on both high-performance and resource-constrained devices, including a Raspberry Pi 4, demonstrating that it delivers sub-second execution times while achieving a constant communication expansion factor of approximately 12x over plaintext size. By directly addressing the communication bottleneck, Hyb-Agg enables scalable, privacy-preserving federated learning that is practical for real-world IoT deployments.

One-Shot Secure Aggregation: A Hybrid Cryptographic Protocol for Private Federated Learning in IoT

TL;DR

<3-5 sentence high-level summary> The paper tackles the prohibitive communication costs of secure aggregation in IoT-enabled federated learning by introducing Hyb-Agg, a one-shot, non-interactive protocol that combines Multi-Key CKKS with ECDH-based additive masking. It proves confidentiality, integrity, and collusion resistance under RLWE and CDH assumptions, modeling the PRF via ChaCha20, and achieves constant per-client communication regardless of participant count with about a 12x expansion. Empirical evaluations on both a high-end machine and a Raspberry Pi 4 demonstrate sub-second end-to-end times on edge hardware and practical scalability for IoT deployments, even under potential server collusion with up to N-2 clients. The work highlights a viable path to scalable, privacy-preserving FL in bandwidth- and energy-constrained environments, while outlining future enhancements for dropout resilience and dynamic participation.

Abstract

Federated Learning (FL) offers a promising approach to collaboratively train machine learning models without centralizing raw data, yet its scalability is often throttled by excessive communication overhead. This challenge is magnified in Internet of Things (IoT) environments, where devices face stringent bandwidth, latency, and energy constraints. Conventional secure aggregation protocols, while essential for protecting model updates, frequently require multiple interaction rounds, large payload sizes, and per-client costs rendering them impractical for many edge deployments. In this work, we present Hyb-Agg, a lightweight and communication-efficient secure aggregation protocol that integrates Multi-Key CKKS (MK-CKKS) homomorphic encryption with Elliptic Curve Diffie-Hellman (ECDH)-based additive masking. Hyb-Agg reduces the secure aggregation process to a single, non-interactive client-to-server transmission per round, ensuring that per-client communication remains constant regardless of the number of participants. This design eliminates partial decryption exchanges, preserves strong privacy under the RLWE, CDH, and random oracle assumptions, and maintains robustness against collusion by the server and up to clients. We implement and evaluate Hyb-Agg on both high-performance and resource-constrained devices, including a Raspberry Pi 4, demonstrating that it delivers sub-second execution times while achieving a constant communication expansion factor of approximately 12x over plaintext size. By directly addressing the communication bottleneck, Hyb-Agg enables scalable, privacy-preserving federated learning that is practical for real-world IoT deployments.

Paper Structure

This paper contains 39 sections, 3 theorems, 5 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Privacy Threats in Federated Learning
  4. Cryptographic Secure Aggregation Protocols
  5. Communication Efficiency in Secure Aggregation for IoT
  6. Preliminaries
  7. Federated Learning
  8. Threat Model and Security Goals
  9. Adversarial Capabilities
  10. Trust Assumptions
  11. Security Goals
  12. Cryptographic Building Blocks
  13. Ring Learning with Errors (RLWE) Assumption
  14. Multi Key Homomorphic Encryption
  15. Elliptic Curve Diffie-Hellman (ECDH)
  16. ...and 24 more sections

Key Result

Theorem 1

Under the RLWE and CDH assumptions, and modeling the ChaCha20 PRG as a pseudorandom function, there is no probabilistic polynomial-time (PPT) adversary, whether server or client, that can distinguish between real protocol executions and simulated executions with random inputs.

Figures (6)

  • Figure 1: Byte Sizes vs. Number of Clients (DataSize = 65536). This figure illustrates how various communication byte sizes per client change as the number of clients increases, while the data size is kept constant at 65536.
  • Figure 2: Byte Sizes vs. Data Size (NumClients = 500). This figure shows how the client uplink, final downlink, and setup byte sizes vary with different data vector sizes, for a fixed number of clients (500)
  • Figure 3: Wall-clock running time per client vs. number of clients for Hyb-Agg. The data vector size is fixed to 65,536.
  • Figure 4: Wall-clock running time per client vs. data vector size for all protocols. The number of clients is fixed at 500.
  • Figure 5: Wall-clock running time for the server vs. number of clients for all protocols. The data vector size is fixed to 65,536.
  • ...and 1 more figures

Theorems & Definitions (5)

  • Theorem 1
  • Lemma 1
  • proof
  • Lemma 2: Collusion Resistance
  • proof