Communication-efficient distributed AI strategies for the IoT edge

Mwase, Christine; Westerlund, Tomi; Tenhunen, Hannu; Zou, Zhuo

doi:10.1016/j.future.2022.01.013

“…Also, ML is being used to improve security, resource allocation, mobility management, and low latency services in MCAs [20]. Distributed ML will be highly important in 6G due to the emerging needs of distributed processing at the edges of the network [45]. Federated learning is currently among the most used distributed ML techniques in communication networks [46], and highly important for 6G due to its capability to be used in distributed manner, much like the the foreseen distributed control nature of 6G networks.…”

Section: Machine Learning Dependability and 6gmentioning

confidence: 99%

Towards dependable 6G networks

Rodríguez¹,

Ahmad²,

Huusko³

et al. 2023

Preprint

5

0

View full text Add to dashboard Cite

<p>6G communication networks must be highly dependable due to the huge number applications, services, and public infrastructures that will be dependant on it. However, the integration of vast number of technologies, ranging from physical layer up to the application layer, and diverse services ranging from entertainment to mission critical public services need investigation of the infrastructure from dependability point of view. We take four dependability concepts such as reliability, availability, safety and security, and discuss the dependability of 6G networks. Since intelligence and distribution of control functions and elements make the core of future networks, specific diligence is given federated learning and edge intelligence. This article, in summary, provides interesting insights into existing challenges and advocates future research through highlighting the most important research directions to make 6G dependable.</p>

show abstract

“…In particular, ML techniques such as deep learning have proved to be extremely efficient in preventing serious security attacks like Distributed Denial of Service (DDoS) attacks [61]. Distributed ML will be highly important in 6G due to the emerging needs of distributed processing at the edges of the network [62]. FL is currently among the most used distributed ML techniques in communication networks [63], and highly important for 6G due to its capability to be used in distributed manner, much like the foreseen distributed control nature of 6G networks.…”

Section: Machine Learning Dependability and 6gmentioning

confidence: 99%

Towards dependable 6G networks

Ahmad¹,

Rodríguez²,

Huusko³

et al. 2023

Preprint

0

View full text Add to dashboard Cite

<p>6G communication networks must be highly dependable due to the huge number applications, services, and public infrastructures that will be dependant on it. However, the integration of vast number of technologies, ranging from physical layer up to the application layer, and diverse services ranging from entertainment to mission critical public services need investigation of the infrastructure from dependability point of view. We take four dependability concepts such as reliability, availability, safety and security, and discuss the dependability of 6G networks. Since intelligence and distribution of control functions and elements make the core of future networks, specific diligence is given federated learning and edge intelligence. This article, in summary, provides interesting insights into existing challenges and advocates future research through highlighting the most important research directions to make 6G dependable.</p>

show abstract

“…In [98], the authors review communication-efficient distributed Machine Learning strategies for the Edge-to-Cloud continuum. They introduce the principles of distributed ML operations and approaches of implementing parallelism and distribution.…”

Section: Related Work and Motivationmentioning

confidence: 99%

“…In summary, all these related works focus on specific domains such as: Machine Learning on the Edge-to-Cloud continuum [7,125,98]; Deep Learning mainly focusing on the Edge, but also discussing hybrid Edge-Cloud deployments [30,150,83,96]; and Data Analytics on Edge-to-Cloud environments [14,131,13].…”

Section: Related Work and Motivationmentioning

confidence: 99%

“…They may become offline at any time for any reason or the network may be congested. This requires fault-tolerant and communication-efficient approaches for distributed model training [80,169,57,98]; and (iii) energy consumption: Edge devices are typically battery-powered, thus they can not handle energy-intensive ML/DL tasks. This should be addressed by energy-efficient inference and training techniques to extend the battery lifetime [50,76].…”

Section: Main Takeawaysmentioning

confidence: 99%

See 1 more Smart Citation