Authors: Yasser H. Khalil, Amir H. Estiri, Mahdi Beitollahi, Nader Asadi, Sobhan Hemati, Xu Li, Guojun Zhang, Xi Chen

Published on: February 02, 2024

Impact Score: 8.07

Arxiv code: Arxiv:2402.01863

Summary

• What is new: A Decentralized Federated Mutual Learning (DFML) framework that supports heterogeneity in models and data without relying on centralized servers or public data.
• Why this is important: Centralized servers in Federated Learning face communication bottlenecks and single points of failure, while existing devices show model and data heterogeneity.
• What the research proposes: DFML, a serverless framework that supports nonrestrictive heterogeneous models and handles model and data heterogeneity through mutual learning without needing public data.
• Results: DFML achieves significant improvements in convergence speed and global accuracy, outperforming baselines with +17.20% and +19.95% increases in global accuracy for the CIFAR-100 dataset with 50 clients under IID and non-IID data shifts, respectively.

Technical Details

Technological frameworks used: Decentralized Federated Mutual Learning (DFML)

Models used: nan

Data used: CIFAR-100 dataset

Potential Impact

Companies specialising in decentralized computing, IoT devices, and mobile networks could benefit. Centralized cloud services might face disruption.

Want to implement this idea in a business?

We have generated a startup concept here: SynapSync.