Matching Pursuit Based Scheduling for Over-the-Air Federated Learning
This work addresses the computational bottleneck in federated learning systems for edge devices, offering an incremental improvement over existing methods.
This paper tackles the problem of device scheduling for over-the-air federated learning by developing low-complexity algorithms using matching pursuit, achieving close-to-optimal performance with significantly reduced computational load compared to benchmarks, as demonstrated on the CIFAR-10 dataset.
This paper develops a class of low-complexity device scheduling algorithms for over-the-air federated learning via the method of matching pursuit. The proposed scheme tracks closely the close-to-optimal performance achieved by difference-of-convex programming, and outperforms significantly the well-known benchmark algorithms based on convex relaxation. Compared to the state-of-the-art, the proposed scheme poses a drastically lower computational load on the system: For $K$ devices and $N$ antennas at the parameter server, the benchmark complexity scales with $\left(N^2+K\right)^3 + N^6$ while the complexity of the proposed scheme scales with $K^p N^q$ for some $0 < p,q \leq 2$. The efficiency of the proposed scheme is confirmed via numerical experiments on the CIFAR-10 dataset.