Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning
Federated learning has emerged as a privacy-preserving technique for collaborative model training across heterogeneously distributed silos. Yet, its reliance on a single central server introduces potential bottlenecks and risks of single-point failure. Decentralizing the server, often referred to as...
Guardado en:
| Autores principales: | , , , , |
|---|---|
| Formato: | Artículo |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | https://arxiv.org/abs/2506.10607 http://arxiv.org/abs/2506.10607 |
| Etiquetas: |
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1879177197530382336 |
|---|---|
| author | Nguyen, Huong Nguyen, Hong-Tri Donta, Praveen Kumar Pirttikangas, Susanna Lovén, Lauri |
| author_browse | Donta, Praveen Kumar Lovén, Lauri Nguyen, Hong-Tri Nguyen, Huong Pirttikangas, Susanna |
| author_facet | Nguyen, Huong Nguyen, Hong-Tri Donta, Praveen Kumar Pirttikangas, Susanna Lovén, Lauri |
| author_sort | Nguyen, Huong |
| building | arXiv |
| collection | Acceso abierto |
| description | Federated learning has emerged as a privacy-preserving technique for collaborative model training across heterogeneously distributed silos. Yet, its reliance on a single central server introduces potential bottlenecks and risks of single-point failure. Decentralizing the server, often referred to as decentralized learning, addresses this problem by distributing the server role across nodes within the network. One drawback regarding this pure decentralization is it introduces communication inefficiencies, which arise from increased message exchanges in large-scale setups. However, existing proposed solutions often fail to simulate the real-world distributed and decentralized environment in their experiments, leading to unreliable performance evaluations and limited applicability in practice. Recognizing the lack from prior works, this work investigates the correlation between model size and network latency, a critical factor in optimizing decentralized learning communication. We propose a graph-based gossiping mechanism, where specifically, minimum spanning tree and graph coloring are used to optimize network structure and scheduling for efficient communication across various network topologies and message capacities. Our approach configures and manages subnetworks on real physical routers and devices and closely models real-world distributed setups. Experimental results demonstrate that our method significantly improves communication, compatible with different topologies and data sizes, reducing bandwidth and transfer time by up to circa 8 and 4.4 times, respectively, compared to naive flooding broadcasting methods. |
| first_indexed | 2026-10-16T00:57:04Z |
| format | Article |
| fullrecord | <oai_dc:dc xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ https://www.openarchives.org/OAI/2.0/oai_dc.xsd"><identifier>arxiv-2506.10607</identifier><datestamp>2025-06-13</datestamp>
<dc:title>Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning</dc:title>
<dc:creator>Nguyen, Huong</dc:creator>
<dc:creator>Nguyen, Hong-Tri</dc:creator>
<dc:creator>Donta, Praveen Kumar</dc:creator>
<dc:creator>Pirttikangas, Susanna</dc:creator>
<dc:creator>Lovén, Lauri</dc:creator>
<dc:subject>Distributed, Parallel, and Cluster Computing</dc:subject>
<dc:description>Federated learning has emerged as a privacy-preserving technique for collaborative model training across heterogeneously distributed silos. Yet, its reliance on a single central server introduces potential bottlenecks and risks of single-point failure. Decentralizing the server, often referred to as decentralized learning, addresses this problem by distributing the server role across nodes within the network. One drawback regarding this pure decentralization is it introduces communication inefficiencies, which arise from increased message exchanges in large-scale setups. However, existing proposed solutions often fail to simulate the real-world distributed and decentralized environment in their experiments, leading to unreliable performance evaluations and limited applicability in practice. Recognizing the lack from prior works, this work investigates the correlation between model size and network latency, a critical factor in optimizing decentralized learning communication. We propose a graph-based gossiping mechanism, where specifically, minimum spanning tree and graph coloring are used to optimize network structure and scheduling for efficient communication across various network topologies and message capacities. Our approach configures and manages subnetworks on real physical routers and devices and closely models real-world distributed setups. Experimental results demonstrate that our method significantly improves communication, compatible with different topologies and data sizes, reducing bandwidth and transfer time by up to circa 8 and 4.4 times, respectively, compared to naive flooding broadcasting methods. </dc:description>
<dc:description>Accepted at 34th International Conference on Computer Communications and Networks (ICCCN 2025)</dc:description>
<dc:date>2025-06-12</dc:date>
<dc:type>text</dc:type>
<dc:identifier>https://arxiv.org/abs/2506.10607</dc:identifier>
<dc:identifier>http://arxiv.org/abs/2506.10607</dc:identifier>
</oai_dc:dc> |
| id | arxiv-2506.10607 |
| institution | arXiv |
| last_indexed | 2026-10-16T00:57:04Z |
| publishDate | 2025 |
| publishDateRange | 2025 |
| publishDateSort | 2025 |
| record_format | oai_dc |
| spelling | arxiv-2506.106072025-06-13 Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning Nguyen, Huong Nguyen, Hong-Tri Donta, Praveen Kumar Pirttikangas, Susanna Lovén, Lauri Distributed, Parallel, and Cluster Computing Federated learning has emerged as a privacy-preserving technique for collaborative model training across heterogeneously distributed silos. Yet, its reliance on a single central server introduces potential bottlenecks and risks of single-point failure. Decentralizing the server, often referred to as decentralized learning, addresses this problem by distributing the server role across nodes within the network. One drawback regarding this pure decentralization is it introduces communication inefficiencies, which arise from increased message exchanges in large-scale setups. However, existing proposed solutions often fail to simulate the real-world distributed and decentralized environment in their experiments, leading to unreliable performance evaluations and limited applicability in practice. Recognizing the lack from prior works, this work investigates the correlation between model size and network latency, a critical factor in optimizing decentralized learning communication. We propose a graph-based gossiping mechanism, where specifically, minimum spanning tree and graph coloring are used to optimize network structure and scheduling for efficient communication across various network topologies and message capacities. Our approach configures and manages subnetworks on real physical routers and devices and closely models real-world distributed setups. Experimental results demonstrate that our method significantly improves communication, compatible with different topologies and data sizes, reducing bandwidth and transfer time by up to circa 8 and 4.4 times, respectively, compared to naive flooding broadcasting methods. Accepted at 34th International Conference on Computer Communications and Networks (ICCCN 2025) 2025-06-12 text https://arxiv.org/abs/2506.10607 http://arxiv.org/abs/2506.10607 |
| spellingShingle | Distributed, Parallel, and Cluster Computing Nguyen, Huong Nguyen, Hong-Tri Donta, Praveen Kumar Pirttikangas, Susanna Lovén, Lauri Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title | Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title_full | Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title_fullStr | Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title_full_unstemmed | Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title_short | Graph-based Gossiping for Communication Efficiency in Decentralized Federated Learning |
| title_sort | graph based gossiping for communication efficiency in decentralized federated learning |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2506.10607 http://arxiv.org/abs/2506.10607 |
| work_keys_str_mv | AT nguyenhuong graphbasedgossipingforcommunicationefficiencyindecentralizedfederatedlearning AT nguyenhongtri graphbasedgossipingforcommunicationefficiencyindecentralizedfederatedlearning AT dontapraveenkumar graphbasedgossipingforcommunicationefficiencyindecentralizedfederatedlearning AT pirttikangassusanna graphbasedgossipingforcommunicationefficiencyindecentralizedfederatedlearning AT lovenlauri graphbasedgossipingforcommunicationefficiencyindecentralizedfederatedlearning |