Age-Based Scheduling Policy for Federated Learning in Mobile Edge Networks

Yang, Howard H.; Arafa, Ahmed; Quek, Tony Q. S.; Poor, H. Vincent

doi:10.1109/icassp40776.2020.9053740

Cited by 162 publications

(79 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Moreover, there are other studies which consider AoI for source-to-destination link in specific scenarios [24][25][26][27][28][29][30]. For instance, the authors in [24] considered an energy harvesting source and optimized the AoI penalty in status update.…”

Section: Related Workmentioning

confidence: 99%

Effective age of information in real-time wireless feedback control systems

Chang

Kizilkaya

et al. 2021

Sci. China Inf. Sci.

View full text Add to dashboard Cite

Ultra-reliable and low-latency communication (URLLC) is one of the most important scenarios in forthcoming fifth generation (5G) cellular networks to ensure timely exchange of information and realize real-time wireless control. In URLLC, timely information update needs to be guaranteed since control performance, e.g., control cost and stability, is directly determined by timely control information update. In this paper, we introduce an effective age of information (EAoI) to evaluate the timeliness of information update in control process. We consider the control process with two phases: sensor to controller phase and controller to actuator phase. We adopt first-generate-first-serve (FGFS) M

show abstract

Section: Related Workmentioning

confidence: 99%

Effective age of information in real-time wireless feedback control systems

Chang

Kizilkaya

et al. 2021

Sci. China Inf. Sci.

View full text Add to dashboard Cite

show abstract

“…The communication burden on the server, generated by the updating of millions of clients, causes significant bottlenecks when scaling up distributed training. To address this communication bottleneck, several techniques, including compression [27,28] and efficient client selection [21,29] have been considered. Various studies have aimed at reducing the communication costs between the clients and server; however, implementing edge computing is the most efficient and practical way to manage numerous clients [13,[30][31][32], as illustrated in Figure 2.…”

Section: Federated Learning Edgementioning

confidence: 99%

Optimal User Selection for High-Performance and Stabilized Energy-Efficient Federated Learning Platforms

et al. 2020

View full text Add to dashboard Cite

Federated learning-enabled edge devices train global models by sharing them while avoiding local data sharing. In federated learning, the sharing of models through communication between several clients and central servers results in various problems such as a high latency and network congestion. Moreover, battery consumption problems caused by local training procedures may impact power-hungry clients. To tackle these issues, federated edge learning (FEEL) applies the network edge technologies of mobile edge computing. In this paper, we propose a novel control algorithm for high-performance and stabilized queue in FEEL system. We consider that the FEEL environment includes the clients transmit data to associated federated edges; these edges then locally update the global model, which is downloaded from the central server via a backhaul. Obtaining greater quantities of local data from the clients facilitates more accurate global model construction; however, this may be harmful in terms of queue stability in the edge, owing to substantial data arrivals from the clients. Therefore, the proposed algorithm varies the number of clients selected for transmission, with the aim of maximizing the time-averaged federated learning accuracy subject to queue stability. Based on this number of clients, the federated edge selects the clients to transmit on the basis of resource status.

show abstract

“…[7,13] maximized the number of selected clients in each communication round. [14] proposed a scheduling policy by jointly accounting for the staleness of the received parameters and the instantaneous channel qualities. Another scheduling criterion is the update significance such as model variance [11] and gradient variance [15].…”

Section: Introductionmentioning

confidence: 99%

Multi-Armed Bandit-Based Client Scheduling for Federated Learning

Xia

Quek

Guo

et al. 2020

IEEE Trans. Wireless Commun.

Self Cite

221

View full text Add to dashboard Cite

By exploiting the computing power and local data of distributed clients, federated learning (FL) features ubiquitous properties such as reduction of communication overhead and preserving data privacy. In each communication round of FL, the clients update local models based on their own data and upload their local updates via wireless channels. However, latency caused by hundreds to thousands of communication rounds remains a bottleneck in FL. To minimize the training latency, this work provides a multi-armed bandit-based framework for online client scheduling (CS) in FL without knowing wireless channel state information and statistical characteristics of clients. Firstly, we propose a CS algorithm based on the upper confidence bound policy (CS-UCB) for ideal scenarios where local datasets of clients are independent and identically distributed (i.i.d.) and balanced. An upper bound of the expected performance regret of the proposed CS-UCB algorithm is provided, which indicates that the regret grows logarithmically over communication rounds. Then, to address non-ideal scenarios with non-i.i.d. and unbalanced properties of local datasets and varying availability of clients, we further propose a CS algorithm based on the UCB policy and virtual queue technique (CS-UCB-Q). An upper bound is also derived, which shows that the expected performance regret of the proposed CS-UCB-Q algorithm can have a sub-linear growth over communication rounds under certain conditions. Besides, the convergence performance of FL training is also analyzed. Finally, simulation results validate the efficiency of the proposed algorithms.

show abstract

Age-Based Scheduling Policy for Federated Learning in Mobile Edge Networks

Cited by 162 publications

References 22 publications

Effective age of information in real-time wireless feedback control systems

Effective age of information in real-time wireless feedback control systems

Optimal User Selection for High-Performance and Stabilized Energy-Efficient Federated Learning Platforms

Multi-Armed Bandit-Based Client Scheduling for Federated Learning

Contact Info

Product

Resources

About