Federated Learning on Non-disjoint Graphs

• Author / Creator

  Tavakoli, Fatemeh

• Federated learning is in widespread use for learning a global model when data is distributed across various distributed clients. In much of the prior work, the data is assumed to consist of independent data points. However, there is often an underlying graph that structures the data points. Such structures emerge in data on social networks, content recommendations, bank transactions data, healthcare data, and other such data where there is a notion of similarity or relation that links data points. Standard federated learning frameworks are not designed specifically for graph data and thus cannot take advantage of graph structure for node classification. We consider federated learning on graph data where a global graph is split among a set of clients, In particular, we consider a non-disjoint split, where there are some nodes that we call anchor nodes, that are present at multiple clients. The learning task is node classification in a semi-supervised scenario where only a small set of nodes have labels. We propose a new federated learning algorithm for non-disjoint graphs that leverages anchor nodes to augment local graph structure for improved node classification. We show through extensive experiments on several graph datasets, that our method outperforms standard methods on the task of node classification.

• Subjects / Keywords

  • Federated learning
  • Graph node classification
  • Graph convolutional networks
  • Semi-supervised learning

• Graduation date

  Fall 2023

• Type of Item

  Thesis

• Degree

  Master of Science

• DOI

  https://doi.org/10.7939/r3-z531-8y15

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.