TITLE:
Federated Graph Learning Based on Graph Distance Computing
AUTHORS:
Wei Gao
KEYWORDS:
Federated Learning, Federated Graph Learning, Graph Distance
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.14 No.11,
November
1,
2024
ABSTRACT: Federated learning is a classic of privacy-preserving learning, which enables collaborative learning without sharing data. Structured data has become the mainstream of current applications, where there is a correlation between samples in the same dataset, which can be represented by a graph. Federated graph learning (FGL) is the integration of structured data into federated learning, assuming that each user has structured graph representation data and uses graph learning models for training. This article proposes an FGL algorithm based on graph distance calculation, which determines the similarity of users in terms of graph distance, then clusters users, and aggregates local models of users in the same cluster. This algorithm can adjust the number of clusters by changing the threshold (the larger the threshold, the fewer clusters and the more users in each cluster, and vice versa), thereby controlling the scope of user collaboration.