Delta sum learning : an approach for fast and global convergence in gossip learning
- Author
- Tom Goethals (UGent) , Merlijn Sebrechts (UGent) , Stijn De Schrijver (UGent) , Filip De Turck (UGent) and Bruno Volckaert (UGent)
- Organization
- Project
- Abstract
- Federated Learning is a popular approach for distributed learning due to its security and computational benefits. With the advent of powerful devices in the network edge, Gossip Learning further decentralizes Federated Learning by removing centralized integration and relying fully on peer to peer updates. However, the averaging methods generally used in both Federated and Gossip Learning are not ideal for model accuracy and global convergence. Additionally, there are few options to deploy Learning workloads in the edge as part of a larger application using a declarative approach such as Kubernetes manifests. This paper proposes Delta Sum Learning as a method to improve the basic aggregation operation in Gossip Learning, and implements it in a decentralized orchestration framework based on Open Application Model, which allows for dynamic node discovery and intent-driven deployment of multiworkload applications. Evaluation results show that Delta Sum performance is on par with alternative integration methods for 10 node topologies, but results in a 58% lower global accuracy drop when scaling to 50 nodes. Overall, it shows strong global convergence and a logarithmic loss of accuracy with increasing topology size compared to a linear loss for alternatives under limited connectivity.
- Keywords
- gossip learning, artificial intelligence, ai, edge, learning, federated learning
Downloads
-
9150 acc.pdf
- full text (Accepted manuscript)
- |
- open access
- |
- |
- 376.38 KB
-
(...).pdf
- full text (Published version)
- |
- UGent only
- |
- |
- 464.41 KB
Citation
Please use this url to cite or link to this publication: http://hdl.handle.net/1854/LU-01M1P0A059NFR914100KWARCX1
- MLA
- Goethals, Tom, et al. “Delta Sum Learning : An Approach for Fast and Global Convergence in Gossip Learning.” NOMS 2026-2026 IEEE Network Operations and Management Symposium, 2026, doi:10.1109/noms69089.2026.11668145.
- APA
- Goethals, T., Sebrechts, M., De Schrijver, S., De Turck, F., & Volckaert, B. (2026). Delta sum learning : an approach for fast and global convergence in gossip learning. NOMS 2026-2026 IEEE Network Operations and Management Symposium. Presented at the IEEE Symposium on Network Operations and Management 2026, Rome, Italy. https://doi.org/10.1109/noms69089.2026.11668145
- Chicago author-date
- Goethals, Tom, Merlijn Sebrechts, Stijn De Schrijver, Filip De Turck, and Bruno Volckaert. 2026. “Delta Sum Learning : An Approach for Fast and Global Convergence in Gossip Learning.” In NOMS 2026-2026 IEEE Network Operations and Management Symposium. https://doi.org/10.1109/noms69089.2026.11668145.
- Chicago author-date (all authors)
- Goethals, Tom, Merlijn Sebrechts, Stijn De Schrijver, Filip De Turck, and Bruno Volckaert. 2026. “Delta Sum Learning : An Approach for Fast and Global Convergence in Gossip Learning.” In NOMS 2026-2026 IEEE Network Operations and Management Symposium. doi:10.1109/noms69089.2026.11668145.
- Vancouver
- 1.Goethals T, Sebrechts M, De Schrijver S, De Turck F, Volckaert B. Delta sum learning : an approach for fast and global convergence in gossip learning. In: NOMS 2026-2026 IEEE Network Operations and Management Symposium. 2026.
- IEEE
- [1]T. Goethals, M. Sebrechts, S. De Schrijver, F. De Turck, and B. Volckaert, “Delta sum learning : an approach for fast and global convergence in gossip learning,” in NOMS 2026-2026 IEEE Network Operations and Management Symposium, Rome, Italy, 2026.
@inproceedings{01M1P0A059NFR914100KWARCX1,
abstract = {{Federated Learning is a popular approach for distributed learning due to its security and computational benefits.
With the advent of powerful devices in the network edge,
Gossip Learning further decentralizes Federated Learning by
removing centralized integration and relying fully on peer to
peer updates. However, the averaging methods generally used in
both Federated and Gossip Learning are not ideal for model
accuracy and global convergence. Additionally, there are few
options to deploy Learning workloads in the edge as part
of a larger application using a declarative approach such as
Kubernetes manifests. This paper proposes Delta Sum Learning
as a method to improve the basic aggregation operation in Gossip
Learning, and implements it in a decentralized orchestration
framework based on Open Application Model, which allows for
dynamic node discovery and intent-driven deployment of multiworkload applications. Evaluation results show that Delta Sum
performance is on par with alternative integration methods for
10 node topologies, but results in a 58% lower global accuracy
drop when scaling to 50 nodes. Overall, it shows strong global
convergence and a logarithmic loss of accuracy with increasing
topology size compared to a linear loss for alternatives under
limited connectivity.}},
author = {{Goethals, Tom and Sebrechts, Merlijn and De Schrijver, Stijn and De Turck, Filip and Volckaert, Bruno}},
booktitle = {{NOMS 2026-2026 IEEE Network Operations and Management Symposium}},
isbn = {{9798331592684}},
issn = {{2374-9709}},
keywords = {{gossip learning,artificial intelligence,ai,edge,learning,federated learning}},
language = {{eng}},
location = {{Rome, Italy}},
pages = {{9}},
title = {{Delta sum learning : an approach for fast and global convergence in gossip learning}},
url = {{http://doi.org/10.1109/noms69089.2026.11668145}},
year = {{2026}},
}
- Altmetric
- View in Altmetric