Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Meta-learning Framework for Tuning Parameters of Protection Mechanisms in Trustworthy Federated Learning

Xiaojin Zhang, Yan Kang, Lixin Fan, Kai Chen, Qiang Yang

TL;DR

This work addresses the challenge of balancing privacy leakage, utility loss, and efficiency reduction in trustworthy federated learning. It develops a meta-learning framework that formulates TFL as a constrained optimization problem and derives bounded measures for the three factors, notably using Bayesian privacy leakage and Jensen–Shannon divergence. The authors provide a practical algorithm to tune protection parameters for four representative mechanisms—Randomization, Paillier Homomorphic Encryption, Secret Sharing, and Compression—together with estimation procedures and error analysis to quantify these parameters in horizontal FL. The framework enables practitioners to select protection settings that meet a privacy budget while minimizing utility loss and efficiency costs, with demonstrated applicability to diverse protection techniques and potential for personalization and multi-objective extensions.

Abstract

Trustworthy Federated Learning (TFL) typically leverages protection mechanisms to guarantee privacy. However, protection mechanisms inevitably introduce utility loss or efficiency reduction while protecting data privacy. Therefore, protection mechanisms and their parameters should be carefully chosen to strike an optimal tradeoff between \textit{privacy leakage}, \textit{utility loss}, and \textit{efficiency reduction}. To this end, federated learning practitioners need tools to measure the three factors and optimize the tradeoff between them to choose the protection mechanism that is most appropriate to the application at hand. Motivated by this requirement, we propose a framework that (1) formulates TFL as a problem of finding a protection mechanism to optimize the tradeoff between privacy leakage, utility loss, and efficiency reduction and (2) formally defines bounded measurements of the three factors. We then propose a meta-learning algorithm to approximate this optimization problem and find optimal protection parameters for representative protection mechanisms, including Randomization, Homomorphic Encryption, Secret Sharing, and Compression. We further design estimation algorithms to quantify these found optimal protection parameters in a practical horizontal federated learning setting and provide a theoretical analysis of the estimation error.

A Meta-learning Framework for Tuning Parameters of Protection Mechanisms in Trustworthy Federated Learning

TL;DR

This work addresses the challenge of balancing privacy leakage, utility loss, and efficiency reduction in trustworthy federated learning. It develops a meta-learning framework that formulates TFL as a constrained optimization problem and derives bounded measures for the three factors, notably using Bayesian privacy leakage and Jensen–Shannon divergence. The authors provide a practical algorithm to tune protection parameters for four representative mechanisms—Randomization, Paillier Homomorphic Encryption, Secret Sharing, and Compression—together with estimation procedures and error analysis to quantify these parameters in horizontal FL. The framework enables practitioners to select protection settings that meet a privacy budget while minimizing utility loss and efficiency costs, with demonstrated applicability to diverse protection techniques and potential for personalization and multi-objective extensions.

Abstract

Trustworthy Federated Learning (TFL) typically leverages protection mechanisms to guarantee privacy. However, protection mechanisms inevitably introduce utility loss or efficiency reduction while protecting data privacy. Therefore, protection mechanisms and their parameters should be carefully chosen to strike an optimal tradeoff between \textit{privacy leakage}, \textit{utility loss}, and \textit{efficiency reduction}. To this end, federated learning practitioners need tools to measure the three factors and optimize the tradeoff between them to choose the protection mechanism that is most appropriate to the application at hand. Motivated by this requirement, we propose a framework that (1) formulates TFL as a problem of finding a protection mechanism to optimize the tradeoff between privacy leakage, utility loss, and efficiency reduction and (2) formally defines bounded measurements of the three factors. We then propose a meta-learning algorithm to approximate this optimization problem and find optimal protection parameters for representative protection mechanisms, including Randomization, Homomorphic Encryption, Secret Sharing, and Compression. We further design estimation algorithms to quantify these found optimal protection parameters in a practical horizontal federated learning setting and provide a theoretical analysis of the estimation error.
Paper Structure (46 sections, 31 theorems, 69 equations, 5 figures, 2 tables, 4 algorithms)

This paper contains 46 sections, 31 theorems, 69 equations, 5 figures, 2 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Privacy Measurements
  4. Privacy Attack and Protection Mechanisms
  5. Trade-off Between Privacy, Utility, and Efficiency
  6. Preliminaries
  7. Notations
  8. Threat Model
  9. Protection Mechanisms
  10. A Framework of Trustworthy Federated Learning
  11. Formulation of Trustworthy Federated Learning
  12. Bayesian Privacy Leakage, Utility Loss and Efficiency Reduction
  13. Bayesian Privacy Leakage
  14. Utility Loss
  15. Efficiency Reduction
  16. ...and 31 more sections

Key Result

Theorem 4.1

We denote $C_{1, k} = \sqrt{{\text{JS}}(F^{\mathcal{O}}_k || F^{\mathcal{B}}_k)}$, $C_1 =\frac{1}{K}\sum_{k=1}^K \sqrt{{\text{JS}}(F^{\mathcal{O}}_k || F^{\mathcal{B}}_k)}$, and denote $C_2 = \frac{1}{2}(e^{2\xi}-1)$, where $\xi = \max_{k\in [K]} \xi_k$, $\xi_k = \max_{w\in \mathcal{W}_k, d \in \mat Specifically, The upper bound for the privacy leakage of federated learning system is Specificall

Figures (5)

  • Figure 1: Outline of this work.
  • Figure 2: Secure FL with Randomization Mechanism
  • Figure 3: Secure FL with Paillier Mechanism
  • Figure 4: Secure FL with Secret Sharing Mechanism
  • Figure 5: Secure FL with Secret Sharing Mechanism

Theorems & Definitions (31)

  • Theorem 4.1
  • Theorem 5.1
  • Theorem 6.1: The Optimal Parameter for Randomization Mechanism
  • Theorem 6.2
  • Theorem 6.3
  • Theorem 6.4
  • Theorem 7.1: The Estimation Error of $C_{1,k}$
  • Theorem 7.2
  • Lemma C.1
  • Theorem C.2: The Estimation Error of $C_{1,k}$
  • ...and 21 more