Reproduce “FedPAE: Peer-Adaptive Ensemble Learning for Asynchronous and Model-Heterogeneous Federated Learning” on OpenResearch

About this paper

Brianna Mueller, W. Nick Street, Stephen Baek, Qihang Lin, Jingyi Yang, Yankun Huang

FedPAE, a fully decentralized personalized federated learning framework, addresses model and data heterogeneity through peer-to-peer model sharing and an adaptive ensemble selection mechanism based on multi-objective optimization. This approach consistently yields higher accuracy, reduced computational cost, and improved protection against negative transfer across diverse non-IID data distributions compared to existing federated learning methods.

Key insights

• Decentralized peer-to-peer model sharing and asynchronous learning effectively eliminate central server dependency and support model heterogeneity in FL.
• Adaptive ensemble selection, optimized for both accuracy and diversity, intrinsically mitigates negative transfer by allowing clients to exclude detrimental peer contributions.
• Balancing local and collective knowledge through selective ensemble formation provides a more nuanced personalization, improving performance in highly heterogeneous environments.

Summary by alphaXiv.