Researchers propose gspDAG-FL, a secure decentralized federated learning framework that derives consensus from the same gossip history used to disseminate model updates. The system allows nodes to exchange payloads only with neighbors while full nodes reconstruct a compact Topology DAG and run Hashgraph-style virtual voting to establish finality over unique model-origin tuples.
- Nodes exchange model payloads exclusively with neighbors, avoiding global coordination costs.
- Full nodes collect event certificates and receiver-endorsed accepted gossip proofs to reconstruct the topology.
- The framework combines payload validation, accepted-proof validation, and private semantic audit before aggregation.
- Experiments on MNIST and Penn Treebank with up to 100 nodes show learning quality close to validation-based ledger FL.
- gspDAG-FL reduces coordination bottlenecks, improves throughput, and maintains high invalid-origin detection under mixed Byzantine and lazy participation.
The authors consider this approach important because it provides provenance finality and resilience to adversarial participants while maintaining the locality benefits of decentralized federated learning.