FedLAM: Low-latency Wireless Federated Learning via Layer-wise Adaptive Modulation

TL;DR

This work tackles high communication latency in wireless federated learning by introducing a layer-wise adaptive modulation scheme that assigns modulation levels to individual DNN layers based on their importance, quantified via Hessian eigenvalues. It extends the FL learning framework with a layer-aware convergence analysis and formulates a practical per-round optimization to maximize loss drop per unit time, employing a power-iteration-based estimate of layer importance and an enumeration (with a group-based reduction) over discrete modulation options. Empirical results on MNIST, Fashion-MNIST, and CIFAR-10 show substantial latency reductions, up to 73.9%, compared to baseline adaptive modulation schemes, while Hessian-eigenvalue computation incurs a modest overhead that is outweighed by the latency savings. The approach provides a scalable pathway to tighten the latency-budget of wireless FL in bandwidth-constrained deployments, with tangible gains in training efficiency and privacy-preserving learning.

Abstract

In wireless federated learning (FL), the clients need to transmit the high-dimensional deep neural network (DNN) parameters through bandwidth-limited channels, which causes the communication latency issue. In this paper, we propose a layer-wise adaptive modulation scheme to save the communication latency. Unlike existing works which assign the same modulation level for all DNN layers, we consider the layers' importance which provides more freedom to save the latency. The proposed scheme can automatically decide the optimal modulation levels for different DNN layers. Experimental results show that the proposed scheme can save up to 73.9% of communication latency compared with the existing schemes.

Figures (3)

• Figure 1: FL over noisy wireless channels.
• Figure 2: Experiment of Fashion-MNIST.
• Figure 3: Experiment of CIFAR-10.