Blockchain-aided wireless federated learning: Resource allocation and client scheduling

TL;DR

This work addresses latency-efficient training in blockchain-aided decentralized federated learning (BDFL) under resource and participation constraints in wireless networks. It introduces DRC-BDFL, a Lyapunov-optimization-based algorithm that jointly optimizes computation, communication, and client scheduling to minimize long-term training delay while respecting energy budgets and participation guarantees. The authors prove a convergence rate of $O(1/T)$ and provide a practical complexity bound, and validate the approach via experiments on SVHN and CIFAR-10, showing significant latency reductions with preserved accuracy compared to baseline schedulers. The results highlight the practical potential of combining decentralized FL with blockchain and Lyapunov methods to realize scalable, secure, and delay-conscious distributed learning in resource-constrained environments.

Abstract

Federated learning (FL) based on the centralized design faces both challenges regarding the trust issue and a single point of failure. To alleviate these issues, blockchain-aided decentralized FL (BDFL) introduces the decentralized network architecture into the FL training process, which can effectively overcome the defects of centralized architecture. However, deploying BDFL in wireless networks usually encounters challenges such as limited bandwidth, computing power, and energy consumption. Driven by these considerations, a dynamic stochastic optimization problem is formulated to minimize the average training delay by jointly optimizing the resource allocation and client selection under the constraints of limited energy budget and client participation. We solve the long-term mixed integer non-linear programming problem by employing the tool of Lyapunov optimization and thereby propose the dynamic resource allocation and client scheduling BDFL (DRC-BDFL) algorithm. Furthermore, we analyze the learning performance of DRC-BDFL and derive an upper bound for convergence regarding the global loss function. Extensive experiments conducted on SVHN and CIFAR-10 datasets demonstrate that DRC-BDFL achieves comparable accuracy to baseline algorithms while significantly reducing the training delay by 9.24% and 12.47%, respectively.

Figures (4)

• Figure 1: Blockchain-aided decentralized federated learning framework
• Figure 2: Average virtual queue backlog with different $V$ values
• Figure 3: Average energy consumption (a) different energy consumption limits; (b) energy consumption of each part ($E^{\max}=0.8\text{J}$)
• Figure 4: Training delay comparison between different algorithms on two datasets. (a) SVHN; (b) CIFAR-10