Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Privacy-Preserving Federated Learning from Partial Decryption Verifiable Threshold Multi-Client Functional Encryption

Minjie Wang, Jinguang Han, Weizhi Meng

TL;DR

The paper tackles the dual challenge of privacy and integrity in federated learning by introducing VTSAFL, a verifiable threshold secure aggregation framework built on partial decryption MCFE. VTSAFL provides aggregation confidentiality with a $t$-of-$s$ threshold and offers verifiability through non-interactive DLEQ proofs, enabling clients to validate results and discard tampered outputs. The authors formalize a Partial Decryption Verifiable Threshold MCFE (VTMCFE), prove IND-security under standard cryptographic assumptions, and integrate it into a scalable FL pipeline with constant-size functional keys and low communication overhead. Empirical evaluation on MNIST and CIFAR-10 demonstrates model performance comparable to existing schemes while achieving substantial reductions in training time (>$40\%$) and communication (up to $50\%$), making it suitable for IoT-scale deployments.

Abstract

In federated learning, multiple parties can cooperate to train the model without directly exchanging their own private data, but the gradient leakage problem still threatens the privacy security and model integrity. Although the existing scheme uses threshold cryptography to mitigate the inference attack, it can not guarantee the verifiability of the aggregation results, making the system vulnerable to the threat of poisoning attack. We construct a partial decryption verifiable threshold multi client function encryption scheme, and apply it to Federated learning to implement the federated learning verifiable threshold security aggregation protocol (VTSAFL). VTSAFL empowers clients to verify aggregation results, concurrently minimizing both computational and communication overhead. The size of the functional key and partial decryption results of the scheme are constant, which provides efficiency guarantee for large-scale deployment. The experimental results on MNIST dataset show that vtsafl can achieve the same accuracy as the existing scheme, while reducing the total training time by more than 40%, and reducing the communication overhead by up to 50%. This efficiency is critical for overcoming the resource constraints inherent in Internet of Things (IoT) devices.

Privacy-Preserving Federated Learning from Partial Decryption Verifiable Threshold Multi-Client Functional Encryption

TL;DR

The paper tackles the dual challenge of privacy and integrity in federated learning by introducing VTSAFL, a verifiable threshold secure aggregation framework built on partial decryption MCFE. VTSAFL provides aggregation confidentiality with a -of- threshold and offers verifiability through non-interactive DLEQ proofs, enabling clients to validate results and discard tampered outputs. The authors formalize a Partial Decryption Verifiable Threshold MCFE (VTMCFE), prove IND-security under standard cryptographic assumptions, and integrate it into a scalable FL pipeline with constant-size functional keys and low communication overhead. Empirical evaluation on MNIST and CIFAR-10 demonstrates model performance comparable to existing schemes while achieving substantial reductions in training time (>) and communication (up to ), making it suitable for IoT-scale deployments.

Abstract

In federated learning, multiple parties can cooperate to train the model without directly exchanging their own private data, but the gradient leakage problem still threatens the privacy security and model integrity. Although the existing scheme uses threshold cryptography to mitigate the inference attack, it can not guarantee the verifiability of the aggregation results, making the system vulnerable to the threat of poisoning attack. We construct a partial decryption verifiable threshold multi client function encryption scheme, and apply it to Federated learning to implement the federated learning verifiable threshold security aggregation protocol (VTSAFL). VTSAFL empowers clients to verify aggregation results, concurrently minimizing both computational and communication overhead. The size of the functional key and partial decryption results of the scheme are constant, which provides efficiency guarantee for large-scale deployment. The experimental results on MNIST dataset show that vtsafl can achieve the same accuracy as the existing scheme, while reducing the total training time by more than 40%, and reducing the communication overhead by up to 50%. This efficiency is critical for overcoming the resource constraints inherent in Internet of Things (IoT) devices.

Paper Structure

This paper contains 28 sections, 1 theorem, 24 equations, 6 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Multi-input Function Encryption
  4. Privacy Preserving Federated Learning
  5. Preliminaries
  6. Notions
  7. Computational Assumptions
  8. Homogeneous linear recursions
  9. Secret Sharing
  10. Discrete Logarithms Equality Protocol
  11. Threat Model
  12. Partial Decryption Verifiable Threshold MCFE
  13. Definition
  14. Security Model
  15. Construction
  16. ...and 13 more sections

Key Result

Theorem 1

We assert that the presented MCFE protocol achieves IND-security within the random oracle model, contingent upon the DDH assumption. Specifically, for a PPT adversary $\mathcal{A}$, the advantage is bounded as follows: where $M$ represents the queries number made to the random oracle $\mathcal{H}$, and $t_{\mathbb{G}}$ denotes the computational cost of a single exponentiation in the group $\mathb

Figures (6)

  • Figure 1: Games $G_{0,1,2}$ and $G_{3. m. \{1,2,3\}}$ for the proof.
  • Figure 2: Games $G_{3.m.\{2,3\}}^*$, $m \in [M]$ for the proof.
  • Figure 3: Our Privacy-Preserving Federated Learning Framework
  • Figure 4: Performance Comparison of Core Cryptographic Primitives
  • Figure 5: Comparison of test accuracy and training loss on MNIST (a, b) and CIFAR-10 (c, d).
  • ...and 1 more figures

Theorems & Definitions (6)

  • Definition 1: DDH Assumption
  • Definition 2: Multi-DDH Assumptionchotard2018decentralized
  • Definition 3: HLR
  • Definition 4: Partial Decryption Verifiable Threshold MCFE
  • Definition 5: IND-Security Game
  • Theorem 1: IND-Security