On the Orchestration of Federated Learning through Vehicular Knowledge Networking

Deveaux, Duncan; Higuchi, Tetsuya; Uçar, Seyhan; Wang, Chang‐Heng; Härri, Jérôme; Altintas, Onur

doi:10.1109/vnc51378.2020.9318386

Cited by 21 publications

(10 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Studies other than optimizing metrics in WFL. Beyond optimization studies, WFL has also been investigated in terms of scheduling policies [28], [29], [30], and mobility-based orchestration [31].…”

Section: A Federated Learning Over Wireless Networkmentioning

confidence: 99%

See 1 more Smart Citation

Joint Optimization of Energy Consumption and Completion Time in Federated Learning

Zhou

Zhao

Han

et al. 2022

2022 IEEE 42nd International Conference on Distributed Computing Systems (ICDCS)

View full text Add to dashboard Cite

Federated Learning (FL) is an intriguing distributed machine learning approach due to its privacy-preserving characteristics. To balance the trade-off between energy and execution latency, and thus accommodate different demands and application scenarios, we formulate an optimization problem to minimize a weighted sum of total energy consumption and completion time through two weight parameters. The optimization variables include bandwidth, transmission power and CPU frequency of each device in the FL system, where all devices are linked to a base station and train a global model collaboratively. Through decomposing the non-convex optimization problem into two subproblems, we devise a resource allocation algorithm to determine the bandwidth allocation, transmission power, and CPU frequency for each participating device. We further present the convergence analysis and computational complexity of the proposed algorithm. Numerical results show that our proposed algorithm not only has better performance at different weight parameters (i.e., different demands) but also outperforms the state of the art.

show abstract

Section: A Federated Learning Over Wireless Networkmentioning

confidence: 99%

“…Simultaneously, it satisfies the system of equations ( 22) and ( 23). Thus, a Newton-like method can be used to update (ν, β), which is done in (31).…”

Section: Solution To Subproblemmentioning

confidence: 99%

Joint Optimization of Energy Consumption and Completion Time in Federated Learning

Zhou

Zhao

Han

et al. 2022

2022 IEEE 42nd International Conference on Distributed Computing Systems (ICDCS)

View full text Add to dashboard Cite

show abstract

“…Beside divergence, heterogeneity can cause unfairness in accuracy across devices, and can produce an un-personalized model. Research works that focused addressing this challenge include [ 87 , 89 , 90 ].…”

Section: Challenges Of Federated Learning and Relevant Research Workmentioning

confidence: 99%

Rebirth of Distributed AI—A Review of eHealth Research

Khan

Alkaabi

2021

Sensors

View full text Add to dashboard Cite

The envisioned smart city domains are expected to rely heavily on artificial intelligence and machine learning (ML) approaches for their operations, where the basic ingredient is data. Privacy of the data and training time have been major roadblocks to achieving the specific goals of each application domain. Policy makers, the research community, and the industrial sector have been putting their efforts into addressing these issues. Federated learning, with its distributed and local training approach, stands out as a potential solution to these challenges. In this article, we discuss the potential interplay of different technologies and AI for achieving the required features of future smart city services. Having discussed a few use-cases for future eHealth, we list design goals and technical requirements of the enabling technologies. The paper confines its focus on federated learning. After providing the tutorial on federated learning, we analyze the Federated Learning research literature. We also highlight the challenges. A solution sketch and high-level research directions may be instrumental in addressing the challenges.

show abstract

“…In this work, we do not presume any connections between vehicles; this positions our work in readily deployable technologies on modern vehicles. A querying approach for vehicle selection was recently studied in [18] in which a request is sent to all available vehicles to detect candidates to participate in Federated Learning. The vehicles send updated responses to the query over time as they are collecting new data, and eventually a subset of the vehicles that answered positively is chosen.…”

Section: Related Workmentioning

confidence: 99%

“…As an example, Federated Learning on vehicles [16] requires the involvement of vehicles that have gathered sufficient suitable data in order not to hinder the learning process [17]. Furthermore exacerbating the issue of vehicle selection are skewed data distributions on vehicles [18], [19] and data minimization directives such as the European GDPR, which dictate to minimize exposure risk and thus overall involvement of customer vehicles. To find vehicles possessing data relevant to an analysis task, one has to overcome the lack of a-priori knowledge about which vehicle has collected which data, without centrally gathering said data first, by leveraging the vehicles' computational power.…”

Section: Introductionmentioning

confidence: 99%

Time- and Computation-Efficient Data Localization at Vehicular Networks’ Edge

et al. 2021

View full text Add to dashboard Cite

As Vehicular Networks rely increasingly on sensed data to enhance functionality and safety, efficient and distributed data analysis is needed to effectively leverage new technologies in real-world applications. Considering the tens of GBs per hour sensed by modern connected vehicles, traditional analysis, based on global data accumulation, can rapidly exhaust the capacity of the underlying network, becoming increasingly costly, slow, or even infeasible. Employing the edge processing paradigm, which aims at alleviating this drawback by leveraging vehicles' computational power, we are the first to study how to localize, efficiently and distributively, relevant data in a vehicular fleet for analysis applications. This is achieved by appropriate methods to spread requests across the fleet, while efficiently balancing the time needed to identify relevant vehicles, and the computational overhead induced on the Vehicular Network. We evaluate our techniques using two large sets of real-world data in a realistic environment where vehicles join or leave the fleet during the distributed data localization process. As we show, our algorithms are both efficient and configurable, outperforming the baseline algorithms by up to a 40× speedup while reducing computational overhead by up to 3×, while providing good estimates for the fraction of vehicles with relevant data and fairly spreading the workload over the fleet. All code as well as detailed instructions are available at https://github.com/dcs-chalmers/dataloc_vn.

show abstract

On the Orchestration of Federated Learning through Vehicular Knowledge Networking

Cited by 21 publications

References 6 publications

Joint Optimization of Energy Consumption and Completion Time in Federated Learning

Joint Optimization of Energy Consumption and Completion Time in Federated Learning

Rebirth of Distributed AI—A Review of eHealth Research

Time- and Computation-Efficient Data Localization at Vehicular Networks’ Edge

Contact Info

Product

Resources

About