Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A New Perspective on Privacy Protection in Federated Learning with Granular-Ball Computing

Guannan Lai, Yihui Feng, Xin Yang, Xiaoyu Deng, Hao Yu, Shuyin Xia, Guoyin Wang, Tianrui Li

TL;DR

This work addresses privacy, efficiency, and utility in Federated Learning from the input perspective. It introduces Granular-Ball Federated Learning (GrBFL), which converts images into graphs through adaptive granular-ball segmentation and uses graph neural networks with proximal-regularized aggregation to protect privacy while preserving performance. Theoretical analysis and experiments on MNIST, CIFAR-10, and CIFAR-100 show that reducing input information content can hinder reconstruction attacks and improve efficiency without sacrificing accuracy, achieving strong PEUM scores. The approach demonstrates practical privacy gains and computational efficiency, offering a new direction for input-aware privacy in collaborative learning systems.

Abstract

Federated Learning (FL) facilitates collaborative model training while prioritizing privacy by avoiding direct data sharing. However, most existing articles attempt to address challenges within the model's internal parameters and corresponding outputs, while neglecting to solve them at the input level. To address this gap, we propose a novel framework called Granular-Ball Federated Learning (GrBFL) for image classification. GrBFL diverges from traditional methods that rely on the finest-grained input data. Instead, it segments images into multiple regions with optimal coarse granularity, which are then reconstructed into a graph structure. We designed a two-dimensional binary search segmentation algorithm based on variance constraints for GrBFL, which effectively removes redundant information while preserving key representative features. Extensive theoretical analysis and experiments demonstrate that GrBFL not only safeguards privacy and enhances efficiency but also maintains robust utility, consistently outperforming other state-of-the-art FL methods. The code is available at https://github.com/AIGNLAI/GrBFL.

A New Perspective on Privacy Protection in Federated Learning with Granular-Ball Computing

TL;DR

This work addresses privacy, efficiency, and utility in Federated Learning from the input perspective. It introduces Granular-Ball Federated Learning (GrBFL), which converts images into graphs through adaptive granular-ball segmentation and uses graph neural networks with proximal-regularized aggregation to protect privacy while preserving performance. Theoretical analysis and experiments on MNIST, CIFAR-10, and CIFAR-100 show that reducing input information content can hinder reconstruction attacks and improve efficiency without sacrificing accuracy, achieving strong PEUM scores. The approach demonstrates practical privacy gains and computational efficiency, offering a new direction for input-aware privacy in collaborative learning systems.

Abstract

Federated Learning (FL) facilitates collaborative model training while prioritizing privacy by avoiding direct data sharing. However, most existing articles attempt to address challenges within the model's internal parameters and corresponding outputs, while neglecting to solve them at the input level. To address this gap, we propose a novel framework called Granular-Ball Federated Learning (GrBFL) for image classification. GrBFL diverges from traditional methods that rely on the finest-grained input data. Instead, it segments images into multiple regions with optimal coarse granularity, which are then reconstructed into a graph structure. We designed a two-dimensional binary search segmentation algorithm based on variance constraints for GrBFL, which effectively removes redundant information while preserving key representative features. Extensive theoretical analysis and experiments demonstrate that GrBFL not only safeguards privacy and enhances efficiency but also maintains robust utility, consistently outperforming other state-of-the-art FL methods. The code is available at https://github.com/AIGNLAI/GrBFL.
Paper Structure (18 sections, 10 equations, 6 figures, 2 tables)

This paper contains 18 sections, 10 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated learning
  4. Granular Ball Computing
  5. Understanding Input in Federated Learning
  6. Problem Definition
  7. Input and Privacy
  8. Input and Utility & Efficiency
  9. Methodology
  10. Overview
  11. Granular-Ball Based Knowledge Reconstruction
  12. Graph Input-Based Aggregations
  13. Complexity analysis.
  14. Experiment
  15. Experiment Setup
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Comparison between the CNNFL and GrBFL processes. The upper section illustrates CNNFL, where image data is directly input into a convolutional neural network, making it vulnerable to attacks. The lower section shows GrBFL, which first reconstructs the image data into a graph before inputting it into a graph neural network, thereby effectively enhancing privacy protection.
  • Figure 2: The Relation Between Input and Privacy
  • Figure 3: The flowchart presents the granular-ball-based knowledge reconstruction algorithm. The upper section provides an overview of the entire process, while the lower section delves into the technical details of the Granular-Rectangle Generation (A) and Graph Construction (B) steps.
  • Figure 4: Privacy Protection Experiment.
  • Figure 5: Comparison of communication efficiency
  • ...and 1 more figures

Theorems & Definitions (1)

  • Definition 1