HybridFLow: SDN-Orchestrated Client Partitioning for Hybrid Federated Learning
HybridFLow uses an SDN controller’s network view to decide which federated-learning clients should train synchronously or asynchronously.
The paper targets cross-silo FL deployments where wide-area communication delays slow each round and worsen stragglers. Its framework estimates per-client communication time before each round, partitions clients accordingly, then feeds measured times back into the controller. In experiments across multiple network topologies, it reached 80% target accuracy 33-40% faster than SmartFLow and cut average round duration by 30-40 seconds. ArXiv · AI/CL/LG's note
The paper targets cross-silo FL deployments where wide-area communication delays slow each round and worsen stragglers. Its framework estimates per-client communication time before each round, partitions clients accordingly, then feeds measured times back into the controller. In experiments across multiple network topologies, it reached 80% target accuracy 33-40% faster than SmartFLow and cut average round duration by 30-40 seconds. ArXiv · AI/CL/LG's note
score 4