🤖 AI Summary
This work addresses the absence of communication topology selection mechanisms tailored to the integration of over-the-air (OTA) computing and decentralized federated learning (DFL). It formulates OTA-DFL for the first time as a distributed query processing problem, establishing an equivalence between this setting and cost-based query optimization. To enable topology-aware cost modeling under privacy constraints, the authors propose a privacy-preserving Count-Min Sketch–based statistical method. By integrating graph-aware cost estimation, topology-aware sparsification, and wireless multi-access superposition computation, the approach achieves near-optimal topologies in 91.4% of evaluated workloads while incurring less than 1.8% additional overhead. The study further provides theoretical guarantees that highly connected topologies exhibit greater robustness to aggressive model compression.
📝 Abstract
Over-the-air (OTA) aggregation exploits the superposition property of wireless multiple-access channels to aggregate model updates from multiple devices within a single transmission slot, significantly reducing communication latency. While OTA computation has been extensively studied for centralized federated learning (FL), its integration with decentralized federated learning (DFL) remains largely unexplored, and principled communication topology selection is absent from existing work.
We present AIRPLAN, a query-optimized topology selection framework for Over-the-Air Decentralized Federated Learning (OTA-DFL). AIRPLAN establishes a formal equivalence between OTA-DFL and distributed query processing, enabling topology selection to be formulated as a cost-based query optimization problem. Using privacy-preserving Count-Min Sketch statistics, AIRPLAN estimates workload characteristics, evaluates a graph-aware cost model across candidate topologies, and selects the communication graph that minimizes training cost while satisfying a target accuracy SLA.
Experiments across five graph families, three vision benchmarks, four client scales, and multiple SNR settings show that AIRPLAN matches the oracle-optimal topology in 91.4% of workloads while introducing less than 1.8% overhead. We further derive theoretical error bounds for topology-aware sparsification, demonstrating that well-connected topologies better tolerate aggressive compression. AIRPLAN introduces a systems-oriented perspective that bridges wireless federated learning and distributed query optimization.