Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedGES: A Federated Learning Approach for BN Structure Learning

Pablo Torrijos, José A. Gámez, José M. Puerta

TL;DR

The paper tackles privacy concerns in Bayesian Network structure learning by introducing FedGES, a federated learning framework that uses constrained GES locally at clients and structural fusion at a server to produce a global BN structure without sharing data or parameters. FedGES preserves the theoretical properties of GES under a distributed, structure-only communication model and employs controlled fusion strategies to promote consensus while limiting complexity. Through extensive experiments on 14 bnlearn BN benchmarks, FedGES demonstrates superior performance over several federated and non-federated baselines, particularly for large or high-dimensional networks and when data are partitioned among many clients. The work offers a practical, privacy-preserving approach to BN structure learning with scalable convergence, reproducible implementation, and clear avenues for future enhancements such as handling non-IID data and adding federated parameter learning.

Abstract

Bayesian Network (BN) structure learning traditionally centralizes data, raising privacy concerns when data is distributed across multiple entities. This research introduces Federated GES (FedGES), a novel Federated Learning approach tailored for BN structure learning in decentralized settings using the Greedy Equivalence Search (GES) algorithm. FedGES uniquely addresses privacy and security challenges by exchanging only evolving network structures, not parameters or data. It realizes collaborative model development, using structural fusion to combine the limited models generated by each client in successive iterations. A controlled structural fusion is also proposed to enhance client consensus when adding any edge. Experimental results on various BNs from {\sf bnlearn}'s BN Repository validate the effectiveness of FedGES, particularly in high-dimensional (a large number of variables) and sparse data scenarios, offering a practical and privacy-preserving solution for real-world BN structure learning.

FedGES: A Federated Learning Approach for BN Structure Learning

TL;DR

The paper tackles privacy concerns in Bayesian Network structure learning by introducing FedGES, a federated learning framework that uses constrained GES locally at clients and structural fusion at a server to produce a global BN structure without sharing data or parameters. FedGES preserves the theoretical properties of GES under a distributed, structure-only communication model and employs controlled fusion strategies to promote consensus while limiting complexity. Through extensive experiments on 14 bnlearn BN benchmarks, FedGES demonstrates superior performance over several federated and non-federated baselines, particularly for large or high-dimensional networks and when data are partitioned among many clients. The work offers a practical, privacy-preserving approach to BN structure learning with scalable convergence, reproducible implementation, and clear avenues for future enhancements such as handling non-IID data and adding federated parameter learning.

Abstract

Bayesian Network (BN) structure learning traditionally centralizes data, raising privacy concerns when data is distributed across multiple entities. This research introduces Federated GES (FedGES), a novel Federated Learning approach tailored for BN structure learning in decentralized settings using the Greedy Equivalence Search (GES) algorithm. FedGES uniquely addresses privacy and security challenges by exchanging only evolving network structures, not parameters or data. It realizes collaborative model development, using structural fusion to combine the limited models generated by each client in successive iterations. A controlled structural fusion is also proposed to enhance client consensus when adding any edge. Experimental results on various BNs from {\sf bnlearn}'s BN Repository validate the effectiveness of FedGES, particularly in high-dimensional (a large number of variables) and sparse data scenarios, offering a practical and privacy-preserving solution for real-world BN structure learning.

Paper Structure

This paper contains 18 sections, 1 equation, 1 figure, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Contributions.
  3. Organization of the paper.
  4. Preliminaries
  5. Bayesian Network
  6. Structure Learning of Bayesian Networks
  7. Greedy Equivalence Search Algorithm
  8. Bayesian Network Structural Fusion
  9. Federated Learning
  10. Related Works
  11. FedGES
  12. Experimental Evaluation
  13. Algorithms
  14. Methodology
  15. Reproducibility
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Mean SMHD of the final global BNs $\mathcal{G}$. Comparison with baselines. "GES Union" lines correspond to running the GES algorithm with no iteration limit on all clients and fusing the resulting networks with a Union. The "Unfinished run" bars correspond to algorithms that cannot finish the execution in a reasonable amount of time.