EdgeDRM learns neighbor-only expert referral for decentralized federated learning
Original title: Learning to Refer: Inference-Time Collaboration in Decentralized Federated Learning
Queries move only between neighboring clients, and only visited clients contribute to the final prediction.
The paper formulates inference-time collaboration in decentralized federated learning as budgeted local expert referral on a client graph. EdgeDRM learns a density-based routing potential at each client and aggregates predictions along the resulting route, using only neighbor-local information and no central router. The authors give estimation guarantees and report that EdgeDRM beats graph-matched decentralized baselines in most settings, with the largest gains under sparse connectivity, on benchmark and real-world medical data.
- Both training and inference stay decentralized.
- Only neighboring clients can receive a referred query.
- Clearest reported gains occur under sparse connectivity.
- Evaluation uses benchmark and real-world medical data.
From the original article
1 min read at arxiv.orgIn decentralized federated learning, heterogeneous clients may develop complementary expertise, yet each query is typically evaluated only by the client where it arrives.
By Yuheng Ma · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.