Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Training Fair Models in Federated Learning without Data Privacy Infringement

Xin Che, Jingdi Hu, Zirui Zhou, Yong Zhang, Lingyang Chu

TL;DR

FedFair is developed, a well-designed federated learning framework, which can successfully train a fair model with high performance without data privacy infringement and uses the fairness estimation to formulate a novel problem of training fair models in federated learning.

Abstract

Training fair machine learning models becomes more and more important. As many powerful models are trained by collaboration among multiple parties, each holding some sensitive data, it is natural to explore the feasibility of training fair models in federated learning so that the fairness of trained models, the data privacy of clients, and the collaboration between clients can be fully respected simultaneously. However, the task of training fair models in federated learning is challenging, since it is far from trivial to estimate the fairness of a model without knowing the private data of the participating parties, which is often constrained by privacy requirements in federated learning. In this paper, we first propose a federated estimation method to accurately estimate the fairness of a model without infringing the data privacy of any party. Then, we use the fairness estimation to formulate a novel problem of training fair models in federated learning. We develop FedFair, a well-designed federated learning framework, which can successfully train a fair model with high performance without data privacy infringement. Our extensive experiments on three real-world data sets demonstrate the excellent fair model training performance of our method.

Training Fair Models in Federated Learning without Data Privacy Infringement

TL;DR

FedFair is developed, a well-designed federated learning framework, which can successfully train a fair model with high performance without data privacy infringement and uses the fairness estimation to formulate a novel problem of training fair models in federated learning.

Abstract

Training fair machine learning models becomes more and more important. As many powerful models are trained by collaboration among multiple parties, each holding some sensitive data, it is natural to explore the feasibility of training fair models in federated learning so that the fairness of trained models, the data privacy of clients, and the collaboration between clients can be fully respected simultaneously. However, the task of training fair models in federated learning is challenging, since it is far from trivial to estimate the fairness of a model without knowing the private data of the participating parties, which is often constrained by privacy requirements in federated learning. In this paper, we first propose a federated estimation method to accurately estimate the fairness of a model without infringing the data privacy of any party. Then, we use the fairness estimation to formulate a novel problem of training fair models in federated learning. We develop FedFair, a well-designed federated learning framework, which can successfully train a fair model with high performance without data privacy infringement. Our extensive experiments on three real-world data sets demonstrate the excellent fair model training performance of our method.

Paper Structure

This paper contains 17 sections, 2 theorems, 28 equations, 2 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Problem Formulation
  4. The DGEO Constraint
  5. Federated Fair Model Training
  6. FedFair: A Practical Solution
  7. Problem Analysis
  8. A Local Estimation Approach
  9. The Federated Estimation Method
  10. Solving the FedFair and LCO Problems
  11. Experimental Results
  12. Experiment Settings
  13. Data Sets
  14. Fairness Measure
  15. Fairness and Accuracy of Trained Models
  16. ...and 2 more sections

Key Result

Theorem 1

Denote by $\sigma_i^2$, $i\in\{1, \ldots, N\}$, the variance of the estimation $\hat{L}_i^{a,c}(\theta) - \hat{L}_i^{b,c}(\theta)$ on the $i$-th client, and by $\sigma_{min}^2 = \min (\sigma_1^2, \ldots, \sigma_N^2)$ the minimum variance. If the data instances in the data set $B_i$ are independently

Figures (2)

  • Figure 1: The testing fairness and testing accuracies of all compared methods. "LR" means that the trained model is a logistic regression model; "NN" means that the trained model is a neural network. "IID" and "non-IID" refer to the IID setting and non-IID setting of data sets, respectively, as described in Section \ref{['sec:data']}.
  • Figure 2: The effect of parameter $\epsilon$ on fairness and accuracy of the logistic regression (LR) models and neural network (NN) models trained by FedFair.

Theorems & Definitions (6)

  • Definition 1
  • Definition 2: Federated Fair Model Training Task
  • Theorem 1
  • proof
  • Theorem 2
  • proof