kth.sePublications KTH
Operational message
There are currently operational disruptions. Troubleshooting is in progress.
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
Open World Learning Graph Convolution for Latency Estimation in Routing Networks
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Theoretical Computer Science, TCS. Ericsson AB, Ericsson Res, Stockholm, Sweden..ORCID iD: 0000-0002-0866-8342
Ericsson AB, Ericsson Res, Stockholm, Sweden..
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Software and Computer systems, SCS.ORCID iD: 0000-0003-4516-7317
KTH, School of Electrical Engineering and Computer Science (EECS), Computer Science, Theoretical Computer Science, TCS.ORCID iD: 0000-0002-5211-112X
2022 (English)In: 2022 INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS (IJCNN), Institute of Electrical and Electronics Engineers (IEEE) , 2022Conference paper, Oral presentation with published abstract (Refereed)
Abstract [en]

Accurate routing network status estimation is a key component in Software Defined Networking. However, existing deep-learning-based methods for modeling network routing are not able to extrapolate towards unseen feature distributions. Nor are they able to handle scaled and drifted network attributes in test sets that include open-world inputs. To deal with these challenges, we propose a novel approach for modeling network routing, using Graph Neural Networks. Our method can also be used for network-latency estimation. Supported by a domainknowledge-assisted graph formulation, our model shares a stable performance across different network sizes and configurations of routing networks, while at the same time being able to extrapolate towards unseen sizes, configurations, and user behavior. We show that our model outperforms most conventional deep-learningbased models, in terms of prediction accuracy, computational resources, inference speed, as well as ability to generalize towards open-world input.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE) , 2022.
Series
IEEE International Joint Conference on Neural Networks (IJCNN), ISSN 2161-4393
Keywords [en]
Graph Convolution, Software Define Networks, Open World Learning
National Category
Computer Sciences
Identifiers
URN: urn:nbn:se:kth:diva-322283DOI: 10.1109/IJCNN55064.2022.9892952ISI: 000867070908062Scopus ID: 2-s2.0-85140761415OAI: oai:DiVA.org:kth-322283DiVA, id: diva2:1717398
Conference
IEEE International Conference on Fuzzy Systems (FUZZ-IEEE) / IEEE World Congress on Computational Intelligence (IEEE WCCI) / International Joint Conference on Neural Networks (IJCNN) / IEEE Congress on Evolutionary Computation (IEEE CEC), JUL 18-23, 2022, Padua, ITALY
Note

Part of proceedings: ISBN 978-1-7281-8671-9

QC 20221219

Available from: 2022-12-08 Created: 2022-12-08 Last updated: 2025-11-10Bibliographically approved
In thesis
1. Generalizable Representation for Wireless Networks Optimization through Native Graph Topology
Open this publication in new window or tab >>Generalizable Representation for Wireless Networks Optimization through Native Graph Topology
2025 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Graph representation learning has become a powerful paradigm for modeling structured data, enabling machine learning systems to reason over relationships, spatial dependencies, and topological patterns. However, its potential in wireless networks remains underexplored, particularly in learning native representations of complex and dynamic wireless environments. This thesis addresses the challenge of applying graph representation learning (such as graph neural networks and transformer architectures) to wireless systems, where topology, domain heuristics, and physical constraints critically impact optimization performance and generalization.

The core problem investigated is how to construct and exploit graph representations that faithfully encode the native structure of wireless networks to enable scalable, topology-aware optimization. This includes coverage relations, interference patterns, and environment-specific propagation effects. Existing solutions in wireless machine learning often overlook these structural priors, resulting in brittle models that generalize poorly across different network deployments.

This thesis introduces a graph-centric methodology to bridge this gap. By representing wireless elements—such as base stations, links, and coverage zones as nodes and their interactions as graph edges, we develop learning architectures that integrate attention mechanisms, domain-aware features, and physics-inspired constraints. Four studies demonstrate the approach across key tasks: routing latency prediction, antenna tilt optimization, real-time radio coverage estimation, and neural ray tracing for link-level modeling.

Our results suggest that these graph-based models significantly outperform traditional baselines, achieving near-simulator accuracy with improved generalization across unseen topologies and user scenarios. They also uncover a correspondence between engineering practices and graph spectral properties, offering a new lens for understanding network design. The proposed methods reduce supervision needs and support scalable deployment across variable network configurations.

Overall, this thesis establishes graph representation learning as a foundational tool for wireless intelligence, enabling structure-informed, optimization-driven modeling across diverse network conditions. These advances pave the way towards future wireless foundation models capable of supporting a wide range of optimization, sensing, and decision-making tasks with minimal retraining.

Abstract [sv]

Inlärning av grafrepresentation har blivit ett kraftfullt paradigm för modellering av strukturerad data, vilket gör det möjligt för maskininlärningssystem att resonera kring relationer, rumsliga beroenden och topologiska mönster. Dess potential inom trådlösa nätverk är dock fortfarande underutforskad, särskilt när det gäller att lära sig nativa representationer av komplexa och dynamiska trådlösa miljöer. Denna avhandling tar upp utmaningen att tillämpa inlärning av grafrepresentation – såsom grafiska neurala nätverk och transformatorarkitekturer – på trådlösa system, där topologi, domänheuristik och fysiska begränsningar kritiskt påverkar optimeringsprestanda och generalisering.

Det centrala problemet som undersöks är hur man konstruerar och utnyttjar grafrepresentationer som troget kodar den nativa strukturen i trådlösa nätverk för att möjliggöra skalbar, topologimedveten optimering. Detta inkluderar optimering av täckningsrelationer, interferensmönster och miljöspecifika utbredningseffekter. Befintliga lösningar inom trådlös maskininlärning förbiser ofta dessa strukturella förutsättningar, vilket resulterar i sköra modeller som generaliserar dåligt över implementeringar och erbjuder begränsad återanvändbarhet.

Denna avhandling introducerar en grafcentrerad metod för att överbrygga detta gap. Genom att representera trådlösa element – såsom basstationer, länkar och täckningszoner – som noder, och deras interaktioner som grafkanter, utvecklar vi inlärningsarkitekturer som integrerar uppmärksamhetsmekanismer, domänmedvetna funktioner och fysikinspirerade begränsningar. Fyra studier demonstrerar denna metod för viktiga trådlösa optimeringsuppgifter: förutsägelse av routinglatens, antennlutningskonfiguration, realtidsuppskattning av radiotäckning och neural strålspårning för modellering på länknivå.

Våra resultat visar att dessa grafbaserade modeller avsevärt överträffar traditionella baslinjer och uppnår nästan simulatornoggrannhet med förbättrad generalisering över osynliga topologier och användarscenarier. De avslöjar också korrespondenser mellan tekniska designmönster och grafspektralegenskaper, vilket erbjuder en ny lins för att förstå och optimera nätverksbeteende. De föreslagna metoderna minskar övervakningsbehovet och stöder skalbar distribution över variabla nätverkskonfigurationer.

Sammantaget etablerar denna avhandling grafrepresentationsinlärning som ett grundläggande verktyg för trådlös intelligens – vilket möjliggör strukturinformerad, optimeringsdriven modellering över olika nätverksförhållanden. Dessa framsteg banar väg för framtida trådlösa grundmodeller som kan stödja ett brett spektrum av uppgifter inom optimering, avkänning och beslutsfattande med minimal omskolning.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2025. p. 58
Series
TRITA-EECS-AVL ; 2025:88
Keywords
Graph neural network, Wireless network, Representation learning, 5G & B5G, Digital twins, Grafiskt neuralt nätverk, Trådlöst nätverk, Representationsinlärning, 5G & B5G, Digitala tvillingar
National Category
Engineering and Technology Electrical Engineering, Electronic Engineering, Information Engineering
Research subject
Computer Science; Electrical Engineering
Identifiers
urn:nbn:se:kth:diva-372589 (URN)978-91-8106-418-6 (ISBN)
Public defence
2025-12-16, Kollegiesalen, Brinellvägen 8, Stockholm, 13:15 (English)
Opponent
Supervisors
Note

QC 20251111

Available from: 2025-11-11 Created: 2025-11-10 Last updated: 2025-12-02Bibliographically approved

Open Access in DiVA

fulltext(1606 kB)283 downloads
File information
File name FULLTEXT01.pdfFile size 1606 kBChecksum SHA-512
900cc4d084bc1170f2444f460ad4aa0b73a20f8b152c5619cdaaa8fdd8d427319fca1f847eede3838afbdaf642e7822ce2cc45932fd6e869b8e9b1d0f9874de7
Type fulltextMimetype application/pdf

Other links

Publisher's full textScopusConference

Authority records

Jin, YifeiGirdzijauskas, SarunasGionis, Aristides

Search in DiVA

By author/editor
Jin, YifeiGirdzijauskas, SarunasGionis, Aristides
By organisation
Theoretical Computer Science, TCSSoftware and Computer systems, SCS
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 283 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

doi
urn-nbn

Altmetric score

doi
urn-nbn
Total: 333 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