kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Blind Asynchronous Over-the-Air Federated Edge Learning
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Network and Systems Engineering.ORCID iD: 0000-0003-4519-9204
KTH, School of Electrical Engineering and Computer Science (EECS), Intelligent systems, Information Science and Engineering.
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Network and Systems Engineering. Princeton University, Department of Electrical and Computer Engineering, New Jersey, USA.ORCID iD: 0000-0002-4503-4242
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Network and Systems Engineering.ORCID iD: 0000-0001-9810-3478
Number of Authors: 42022 (English)In: 2022 IEEE GLOBECOM Workshops, GC Wkshps 2022: Proceedings, Institute of Electrical and Electronics Engineers (IEEE) , 2022, p. 1834-1839Conference paper, Published paper (Refereed)
Abstract [en]

Federated Edge Learning (FEEL) is a distributed machine learning technique where each device contributes to training a global inference model by independently performing local computations with their data. More recently, FEEL has been merged with over-the-air computation (OAC), where the global model is calculated over the air by leveraging the superposition of analog signals. However, when implementing FEEL with OAC, there is the challenge on how to precode the analog signals to overcome any time misalignment at the receiver. In this work, we propose a novel synchronization-free method to recover the parameters of the global model over the air without requiring any prior information about the time misalignments. For that, we construct a convex optimization based on the norm minimization problem to directly recover the global model by solving a convex semi-definite program. The performance of the proposed method is evaluated in terms of accuracy and convergence via numerical experiments. We show that our proposed algorithm is close to the ideal synchronized scenario by 10%, and performs 4times better than the simple case where no recovering method is used.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE) , 2022. p. 1834-1839
Keywords [en]
Asynchronous, federated edge learning, over-the-air computation, time misalignment
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:kth:diva-333461DOI: 10.1109/GCWkshps56602.2022.10008588Scopus ID: 2-s2.0-85146155208OAI: oai:DiVA.org:kth-333461DiVA, id: diva2:1785375
Conference
2022 IEEE GLOBECOM Workshops, GC Wkshps 2022, Virtual, Online, Brazil, Dec 4 2022 - Dec 8 2022
Note

Part of ISBN 9781665459754

QC 20230802

Available from: 2023-08-02 Created: 2023-08-02 Last updated: 2023-11-01Bibliographically approved
In thesis
1. ChannelComp: A general framework for computing by digital communication
Open this publication in new window or tab >>ChannelComp: A general framework for computing by digital communication
2023 (English)Licentiate thesis, comprehensive summary (Other academic)
Abstract [en]

The imminent Internet of Things, fueled by 6G networks and machine learning technologies, is set to shift wireless communication to machine-centric paradigms, revolutionizing sectors such as healthcare or industrial automation through efficient data handling. However, this connectivity boom poses challenges, including straining existing communication systems due to increased data traffic and computational demands.

Over-the-air computation (OAC) presents a feasible solution, allowing the summation of transmitted signals at a common receiver through analog amplitude modulation. Designed to enable concurrent data collection and computation at the network edge, OAC seeks to lessen the central system burden, reducing latency and energy usage while enabling real-time analytics. This approach is particularly beneficial for federated learning, a machine learning technique that operates across decentralized devices. However, OAC's dependence on analog communication poses notable challenges, including signal distortion during transmission and the limited availability of devices supporting analog modulations. Digital modulation is a preferable alternative, recognized for its excellent channel correction capabilities and broad acceptance in modern wireless devices. Nevertheless, its integration into OAC is perceived as a significant hurdle, with overlapping digitally modulated signals threatening the fundamental concept of simultaneous data collection and computation.

The first part of the thesis provides an overview of communication systems, specifically focusing on the relevant OAC methodologies for analog and digital parts and their application in ML, particularly in training federated learning models. Subsequently, an exhaustive literature review concerning analog OAC techniques is undertaken, identifying existing limitations within this domain. The central thrust of our research is then introduced, proposing an innovative digital OAC approach along with a fresh perspective on the communication systems models designed for executing the computation. The chapter concludes with a summary of the principal contributions of each paper included within the thesis.

In the second part, we introduce ChannelComp, a groundbreaking computing approach compatible with current digital communication systems, including smartphones and IoT devices. A detailed analysis of ChannelComp's functions reveals how it enables digital modulation schemes to perform computations, addressing a critical gap in previous research. Moreover, introducing pre-coders designed for function computation over the multiple access channel, combined with a feasibility optimization problem framework, allows for seamless integration with current systems. Compared to OAC, restricted to analog modulations, ChannelComp exhibits broader computational capabilities and adherence to strict computation time constraints, thus showcasing its robust potential for future massive machine-type communications. This innovative method signifies a promising direction toward sustainable and efficient future wireless communication.

Abstract [sv]

Den nära förestående Internet of Things, drivet av 6G-nätverk och maskininlärningsteknologier, är på väg att förändra trådlös kommunikation till maskincentrerade paradigm, revolutionerande sektorer som hälso- och sjukvård samt industriell automatisering genom effektiv datahantering. Dock medför denna uppkopplingsboom utmaningar, inklusive påfrestningar på befintliga kommunikationssystem på grund av ökad datatrafik och beräkningsbehov.

Over-the-air-beräkning (OAC) framstår som en genomförbar lösning, genom att tillåta summering av överförda signaler hos en gemensam mottagare genom analog amplitudmodulering. Utformad för att möjliggöra samtidig datainsamling och beräkning vid nätverkskanten, strävar OAC efter att minska den centrala systembelastningen, minska latens och energiförbrukning samtidigt som det möjliggör realtidsanalys. Denna metod är särskilt fördelaktig för federerad inlärning, en maskininlärningsteknik som fungerar över decentraliserade enheter. Dock medför OAC:s beroende av analog kommunikation märkbara utmaningar, inklusive signal distortion under överföring och begränsad tillgänglighet av enheter som stöder analoga moduleringar. Digital modulering är ett föredraget alternativ, erkänt för dess utmärkta kanalkorrigeringsegenskaper och bred acceptans i moderna trådlösa enheter. Trots detta uppfattas dess integration i OAC som ett betydande hinder, med överlappande digitalt modulerade signaler som hotar den grundläggande konceptet med samtidig datainsamling och beräkning.

Den första delen av avhandlingen ger en översikt över kommunikationssystem, med särskilt fokus på relevanta OAC-metodiker för analoga och digitala delar och deras tillämpning i ML, särskilt vid träning av federerade inlärningsmodeller. Därefter genomförs en omfattande litteraturöversikt angående analoga OAC-tekniker, där befintliga begränsningar inom detta område identifieras. Forskningens centrala drivkraft introduceras sedan, med förslag på en innovativ digital OAC-metod tillsammans med ett nytt perspektiv på kommunikationssystemmodeller utformade för att utföra beräkningen. Kapitlet avslutas med en sammanfattning av de huvudsakliga bidragen från varje artikel inkluderad i avhandlingen.

I den andra delen introducerar vi ChannelComp, en ny och banbrytande beräkningsmetod som är kompatibel med nuvarande digitala kommunikationssystem, inklusive smartphones och IoT-enheter. En detaljerad analys av ChannelComp:s funktioner avslöjar hur den möjliggör digitala moduleringsscheman för att utföra beräkningar, vilket adresserar en kritisk lucka i tidigare forskning. Dessutom möjliggör introduktionen av förkodare utformade för funktionsberäkning över den fleraccessa kanalen, kombinerat med ett ramverk för genomförbarhetsoptimeringsproblem, en sömlös integration med nuvarande system. Jämfört med OAC, begränsad till analoga moduleringar, uppvisar ChannelComp bredare beräkningsmöjligheter och efterlevnad av strikta beräkningstidsbegränsningar, vilket visar dess robusta potential för framtida massiva maskintypkommunikationer. Denna innovativa metod signalerar en lovande riktning mot hållbar och effektiv framtida trådlös kommunikation.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2023. p. 161
Series
TRITA-EECS-AVL ; 2023:75
National Category
Communication Systems
Research subject
Telecommunication
Identifiers
urn:nbn:se:kth:diva-338940 (URN)978-91-8040-741-0 (ISBN)
Presentation
2023-11-20, E32 https://kth-se.zoom.us/j/61848116543, Osquars backe 2, E-huset, huvudbyggnaden, Lindstedtsvägen 3, floor 3, Stockholm, 13:00 (English)
Opponent
Supervisors
Note

QC 20231031

Available from: 2023-10-31 Created: 2023-10-31 Last updated: 2023-11-13Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Razavikia, SaeedPeris, Jaume AngueraBarros da Silva Jr., José MairtonFischione, Carlo

Search in DiVA

By author/editor
Razavikia, SaeedPeris, Jaume AngueraBarros da Silva Jr., José MairtonFischione, Carlo
By organisation
Network and Systems EngineeringInformation Science and Engineering
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 44 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf