Special Topic on Federated Learning over Wireless Networks

Federated learning has revolutionized the way we ap‐proach machine learning by enabling multiple edge de‐vices to collaboratively learn a shared machine learn‐ing model without the need for centralized data collec‐tion.Such a new machine learning paradigm has gained sig‐nificant attention in recent years due to its ability to address privacy and security concerns associated with centralized learning,as well as its potential to reduce communication overhead and improve scalability.Deploying cross-device fed‐erated learning at the network edge over wireless networks has further extended its potential due to the close proximity to the gigantic number of mobile data and computing power provided by the surging number of Internet of Things (IoT) devices,and is expected to breed new intelligent applications that demand delay-sensitive and mission-critical services,such as smart in‐dustry,auto-driving,and metaverse.Despite its great promise,the successful deployment of federated learning over wireless networks has also presented its own unique set of challenges,including network heterogeneity,communication delays,and unreliable connections.

In this special issue,a series of articles are presented to ad‐dress the aforementioned challenges and propose innovative solutions to enabling federated learning over wireless net‐works.These articles cover a wide range of topics,including wireless communication protocols,optimization algorithms,se‐curity and privacy concerns,network architecture designs,and the application of federated learning in IoT and 5G net‐works.The call-for-papers of this special issue have brought excellent submissions in both quality and quantity.After tworound reviews,five excellent papers have been selected for publication in this special issue which is organized as follows.

The first paper titled “Adaptive Retransmission Design for Wireless Federated Edge Learning” proposes a novel retrans‐mission scheme for wireless federated edge learning (FEEL).The conventional retransmission schemes for wireless sys‐tems,which aim to maximize the system throughput or mini‐mize the packet error rate,are not suitable for the FEEL sys‐tem.The proposed scheme makes a tradeoff between model training accuracy and retransmission latency,with a retrans‐mission device selection criterion designed based on the chan‐nel condition,the number of local data,and the importance of model update.Additionally,the air interface signaling is de‐signed to facilitate the implementation of the proposed scheme in practical scenarios.Simulation experiments validate the ef‐fectiveness of the proposed retransmission scheme.

The second paper titled “Reliable and Privacy-Preserving Federated Learning with Anomalous Users” proposes a reliable and privacy-preserving federated learning scheme named RPPFL,based on a single-cloud model.The scheme addresses the issue of anomalous users holding low-quality data,which may reduce the accuracy of trained models.The proposed ap‐proach identifies the user’s reliability and thereby decreases the impact of anomalous users,based on the truth discovery technique.The additively homomorphic cryptosystem is utilized to provide comprehensive privacy preservation (user’s local gra‐dient privacy and reliability privacy).Rigorous theoretical analy‐sis shows the security of RPPFL,and extensive experiments based on open datasets demonstrate that RPPEL compares favor‐ably with existing works in terms of efficiency and accuracy.

The third paper titled “RIS-Assisted Federated Learning in Multi-Cell Wireless Networks” proposes a reconfigurable in‐telligent surface (RIS)-assisted AirComp-based federated learning (FL) in multi-cell networks.The proposed system en‐hances the poor user signal caused by channel fading,espe‐cially for the device at the cell edge,and reduces inter-cell in‐terference.The convergence of FL in the proposed system is analyzed,and the optimality gap for FL is derived.To mini‐mize the optimality gap,the paper formulates a joint uplink and downlink optimization problem,which is then divided into two separable nonconvex subproblems.Following the succes‐sive convex approximation (SCA) method,the paper first ap‐proximates the nonconvex term to a linear form,and then alter‐nately optimizes the beamforming vector and phase-shift ma‐trix for each cell.Simulation results demonstrate the advan‐tages of deploying a RIS in multi-cell networks,and the pro‐posed system significantly improves the performance of FL.

The fourth paper titled “Hierarchical Federated Learning: Architecture,Challenges,and Its Implementation in Vehicu‐lar Networks” discusses hierarchical federated learning (HFL) and its implementation in vehicular networks.HFL,with a cloud-edge-client hierarchy,can leverage the large coverage of cloud servers and the low transmission latency of the edge server.The limited number of participants in vehicular net‐works and vehicle mobility degrades the performance of FL training,and HFL is promising in vehicular networks due to its lower latency,wider coverage,and more participants.The paper clarifies new issues in HFL,reviews several existing so‐lutions,introduces some typical use cases in vehicular net‐works,and discusses the initial efforts on implementing HFL in vehicular networks.

The fifth paper titled “Secure Federated Learning over Wireless Communication Networks with Model Compression” addresses the vulnerability of FL to gradient leakage attacks.A method is proposed to compress the model size to reduce the leakage risk and enhance the efficiency of FL.Specifically,this paper presents a new scheme that applies low-rank matrix approximation to compress the model and uses a secure matrix factorization technique to recover the original model.Experi‐ments showed that the proposed method achieved better accu‐racy and security compared with the state-of-the-art methods.

To conclude,it is hoped that this special issue will serve as a valuable resource for researchers,practitioners,and students who are interested in federated learning over wireless networks.We also hope that it will inspire further research in this field,leading to new and innovative solutions that will drive the evolu‐tion of machine learning.Finally,we would like to express our sincere gratitude to all the authors,reviewers,and editorial staff who have contributed to the success of this special issue.Hope‐fully,the articles in this special issue are both insightful and in‐formative for prospective readers in the field.