Research Tools & Code·arXiv cs.LG·14h ago

Federated learning achieves consensus without central coordination

Illustration accompanying: Secure Decentralized Federated Learning via Gossip and Virtual Voting

Federated learning systems face a fundamental tension: removing central servers improves privacy and resilience, but sacrifices auditability and consensus finality. This paper introduces gspDAG-FL, which resolves that tradeoff by anchoring consensus directly to the gossip protocol that already distributes model updates across peers. Rather than layering blockchain infrastructure on top of decentralized training, the framework reconstructs a compact DAG from peer endorsements and applies Hashgraph-style consensus to the same communication graph used for model dissemination. The result is a system that maintains locality and Byzantine fault tolerance without reintroducing global coordination bottlenecks, addressing a critical pain point for large-scale distributed ML.

Modelwire context

Explainer

The key innovation is embedding consensus finality directly into the gossip protocol itself rather than bolting on a separate blockchain layer. Most decentralized federated learning systems treat consensus and model distribution as separate problems; gspDAG-FL reconstructs a DAG from the same peer endorsements that already carry model updates, collapsing two communication graphs into one.

This work sits at the intersection of two recent Modelwire themes: privacy-preserving distributed systems and the tension between decentralization and auditability. The EdgeRefine paper (July 9) tackled privacy-utility tradeoffs in graph structures by refining rather than brute-forcing noise; gspDAG-FL takes a similar philosophy toward the consensus-resilience tradeoff, avoiding the typical move of layering additional infrastructure. Both papers signal a shift from accepting fundamental constraints to engineering around them. The MPFlow work on Lightning Network optimization (same date) also operates within real network topology rather than idealized assumptions, suggesting a broader maturation in how researchers approach decentralized systems.

If gspDAG-FL implementations achieve Byzantine fault tolerance with fewer than 4f+1 total peers (the standard lower bound for classical BFT) on realistic federated learning workloads with 100+ participants, that confirms the gossip-DAG approach genuinely reduces coordination overhead. Otherwise, watch whether the paper's experiments stay confined to small-scale synthetic networks or scale to production federated settings like cross-silo healthcare or finance deployments within 18 months.

Coverage we drew on

EdgeRefine: Privacy-Utility Balance for Graphs via Jaccard Sampling under Edge Differential Privacy · arXiv cs.LG

This analysis is generated by Modelwire’s editorial layer from our archive and the summary above. It is not a substitute for the original reporting. How we write it.

MentionsgspDAG-FL · Hashgraph · federated learning · Byzantine fault tolerance

Read full story at arXiv cs.LG →(arxiv.org)

Modelwire Editorial

This synthesis and analysis was prepared by the Modelwire editorial team. We use advanced language models to read, ground, and connect the day’s most significant AI developments, providing original strategic context that helps practitioners and leaders stay ahead of the frontier.

Our mission How we write

Modelwire summarizes, we don’t republish. arXiv cs.LG originally reported this story as “Secure Decentralized Federated Learning via Gossip and Virtual Voting”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.