Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A New Perspective to Boost Performance Fairness for Medical Federated Learning

Yunlu Yan, Lei Zhu, Yuexiang Li, Xinxing Xu, Rick Siow Mong Goh, Yong Liu, Salman Khan, Chun-Mei Feng

TL;DR

Fed-LWR is proposed to improve performance fairness from the perspective of feature shift, a key issue influencing the performance of medical FL systems caused by domain shift, and dynamically perceive the bias of the global model across all hospitals.

Abstract

Improving the fairness of federated learning (FL) benefits healthy and sustainable collaboration, especially for medical applications. However, existing fair FL methods ignore the specific characteristics of medical FL applications, i.e., domain shift among the datasets from different hospitals. In this work, we propose Fed-LWR to improve performance fairness from the perspective of feature shift, a key issue influencing the performance of medical FL systems caused by domain shift. Specifically, we dynamically perceive the bias of the global model across all hospitals by estimating the layer-wise difference in feature representations between local and global models. To minimize global divergence, we assign higher weights to hospitals with larger differences. The estimated client weights help us to re-aggregate the local models per layer to obtain a fairer global model. We evaluate our method on two widely used federated medical image segmentation benchmarks. The results demonstrate that our method achieves better and fairer performance compared with several state-of-the-art fair FL methods.

A New Perspective to Boost Performance Fairness for Medical Federated Learning

TL;DR

Fed-LWR is proposed to improve performance fairness from the perspective of feature shift, a key issue influencing the performance of medical FL systems caused by domain shift, and dynamically perceive the bias of the global model across all hospitals.

Abstract

Improving the fairness of federated learning (FL) benefits healthy and sustainable collaboration, especially for medical applications. However, existing fair FL methods ignore the specific characteristics of medical FL applications, i.e., domain shift among the datasets from different hospitals. In this work, we propose Fed-LWR to improve performance fairness from the perspective of feature shift, a key issue influencing the performance of medical FL systems caused by domain shift. Specifically, we dynamically perceive the bias of the global model across all hospitals by estimating the layer-wise difference in feature representations between local and global models. To minimize global divergence, we assign higher weights to hospitals with larger differences. The estimated client weights help us to re-aggregate the local models per layer to obtain a fairer global model. We evaluate our method on two widely used federated medical image segmentation benchmarks. The results demonstrate that our method achieves better and fairer performance compared with several state-of-the-art fair FL methods.

Paper Structure

This paper contains 11 sections, 6 equations, 2 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Method
  3. Preliminaries
  4. Fed-LWR: Layer-wise Re-weighting for Federated Learning
  5. Experiment
  6. Experimental Setup
  7. Comparison with State-of-the-Arts
  8. Analytical Studies
  9. Conclusion
  10. Acknowledgments.
  11. Disclosure of Interests.

Figures (2)

  • Figure 1: (a) Overview ofFed-LWR. During the parameter aggregation stage of $t$-th round, Fed-LWR calculate layer-wise CKA similarity $\delta_k = \{\delta_k^1, \delta_k^2\ldots,\delta_k^M\}$ between the local model $\boldsymbol{w}_k^t$ and anchor $\boldsymbol{w}_G^t$ averaged by server on hospital $k$. The CKA scores will be used to re-aggregate the local models to obtain fair global model $\boldsymbol{\hat{w}}_G^t$. (b) Variations in CKA similarity versus the layers of local models from two pairs of clients, which are randomly selected. This reveals that the differences between models vary with different layers.
  • Figure 2: (a) Avg. and (b) Std. versus the number of communication rounds on the testing set of ProstateMRI liu2020ms.

Theorems & Definitions (1)

  • definition thmcounterdefinition