Adaptive Decentralized Federated Learning in Energy and Latency Constrained Wireless Networks

TL;DR

This work tackles the challenge of deploying decentralized federated learning (DFL) in energy- and latency-constrained wireless networks with device heterogeneity. It develops a convergence-bound framework to quantify how local training rounds and energy-aware, graph-based aggregations affect performance, and then derives closed-form solutions for per-device training allocations and energy-saving aggregation (MST when channel information is stable, Ring-AllReduce otherwise). The proposed Algorithm 3 orchestrates adaptive local training across iterations and devices and selects energy-efficient aggregation schemes, yielding better model accuracy and reduced energy consumption compared to fixed-round schemes and baseline gossip methods. Overall, the paper provides a principled, scalable approach to balancing computation, communication, and energy budgets in practical DFL deployments.

Abstract

In Federated Learning (FL), with parameter aggregated by a central node, the communication overhead is a substantial concern. To circumvent this limitation and alleviate the single point of failure within the FL framework, recent studies have introduced Decentralized Federated Learning (DFL) as a viable alternative. Considering the device heterogeneity, and energy cost associated with parameter aggregation, in this paper, the problem on how to efficiently leverage the limited resources available to enhance the model performance is investigated. Specifically, we formulate a problem that minimizes the loss function of DFL while considering energy and latency constraints. The proposed solution involves optimizing the number of local training rounds across diverse devices with varying resource budgets. To make this problem tractable, we first analyze the convergence of DFL with edge devices with different rounds of local training. The derived convergence bound reveals the impact of the rounds of local training on the model performance. Then, based on the derived bound, the closed-form solutions of rounds of local training in different devices are obtained. Meanwhile, since the solutions require the energy cost of aggregation as low as possible, we modify different graph-based aggregation schemes to solve this energy consumption minimization problem, which can be applied to different communication scenarios. Finally, a DFL framework which jointly considers the optimized rounds of local training and the energy-saving aggregation scheme is proposed. Simulation results show that, the proposed algorithm achieves a better performance than the conventional schemes with fixed rounds of local training, and consumes less energy than other traditional aggregation schemes.

Figures (9)

• Figure 1: Framework of the iterative algorithm of DFL.
• Figure 2: An example of the aggregation scheme based on MST (After the device 3 obtaining the aggregated parameters, it transmits this result to others on the same link).
• Figure 3: An example of the aggregation scheme based on Ring-AllReduce.
• Figure 4: Simulations on Fashion-MNIST and CIFAR-10 datasets with MLP and CNN under different aggregation schemes and topolgies. (a) Test accuracy with iterations of MLP with Fashion-MNIST on the grid topology. (b) Test accuracy with iterations of CNN with CIFAR-10 on the ring topology.
• Figure 5: Simulations on MNIST and Fashion-MNIST datasets with MLP with different number of device in non-i.i.d. case. (a) Test accuracy with iterations of MLP with MNIST on different $N$. (b) Test accuracy with iterations of MLP with Fashion-MNIST on different $N$.