kth.sePublications KTH
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
Accelerated ADMM Variants for Distributed Optimization: Algorithms for Dynamic and Large Networks
KTH, School of Electrical Engineering and Computer Science (EECS), Information Science and Engineering.ORCID iD: 0000-0002-2439-2884
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

This doctoral thesis presents a comprehensive summary of research efforts with the aim of advancing the state-of-the-art in decentralized optimization. As modern distributed systems grow in scale and complexity, traditional optimization methods face significant bottlenecks. This work, presented as a compilation of five papers, specifically targets the Alternating Direction Method of Multipliers (ADMM). The central objective is to re-engineer ADMM to overcome the dual challenges of convergence latency and communication inefficiencies in peer-to-peer networks.  The first part of the thesis focuses on decentralization, convergence speed, and computational cost. While centralized ADMM algorithms struggle with scalability and bottlenecks, existing decentralized schemes rely on either excessive computations and messaging or completely sequential operations, causing a prolonged time required to reach convergence. To tackle these problems, we introduce two fast-converging decentralized ADMM algorithms. Our theoretical analysis confirms that our algorithms retain the classical convergence properties of centralized ADMM while maintaining a low per-node complexity of $O(1)$. Numerical simulations further demonstrate that our algorithms converge significantly faster than state-of-the-art decentralized implementations, providing clear conditions under which they outperform traditional benchmarks.  The second part of the thesis addresses the straggler problem, which arises from the conventional ADMM requirement for global synchronization, where faster nodes are forced to remain idle until the slowest nodes complete their local updates, impeding the progress of the entire network. Here, we introduce three algorithms to allow the system to remain productive even under single-point-of-failure scenarios or extreme hardware variance. The first algorithm achieves straggler-resilience by allowing the nodes to proceed to the next iteration even when one or more nodes have not provided an update for one or more iterations. The second algorithm is a decentralized version of the first algorithm. The second algorithm enforces fast convergence as well as robustness against stragglers and single points of failure through decentralized, asynchronous, and concurrent operations. The third algorithm extends the second algorithm by enforcing robustness against uncertainties with the help of a time-tracking scheme. Through theoretical analyses, we establish the convergence properties of our algorithms and show that our decentralized algorithms achieve a computational complexity of $O(1)$ for each worker node, whereas our centralized algorithm achieves a computational complexity of $O(N)$ for the central node and $O(1)$ for each of the remaining nodes. Through numerical simulations with various settings, we show that our algorithms have converged significantly faster than several state-of-the-art ADMM algorithms with well-established convergence properties.

The final part of the thesis extends these optimizations to highly volatile systems characterized by message dropouts and dynamic topologies. Here, we introduce two algorithms to adapt ADMM to dynamic systems, where new nodes may be added amidst the process at the same time as the system may encounter issues with stragglers and message dropouts. The algorithms achieve fast convergence and flexibility by allowing nodes to choose a step size that is best suited for their own system and by allowing the nodes to move on to the next iteration even when not all nodes have made an update for the current iteration. More importantly, these algorithms incorporate a contribution tracking mechanism to ensure consistency despite message loss. Furthermore, these algorithms enforce robustness against uncertainties by removing the need to predefine the waiting time or the minimum number of updates before moving to the next iteration. Here, an approximation mechanism is also introduced to give stragglers more time to compute their variables while the faster nodes move on to the next iteration.

To summarize, this thesis provides accelerated algorithms for fast convergence in distributed optimization, with solutions tailored for both large-scale and dynamic networks.

Abstract [sv]

Denna doktorsavhandling presenterar en omfattande sammanfattning av forskningsinsatser som syftar till att främja den senaste tekniken inom decentraliserad optimering. I takt med att moderna distribuerade system växer i skala och komplexitet möter traditionella optimeringsmetoder betydande flaskhalsar. Detta arbete, som presenteras som en sammanställning av fem artiklar, riktar sig specifikt mot Alternating Direction Method of Multipliers (ADMM). Det centrala målet är att omkonstruera ADMM för att övervinna de dubbla utmaningarna med konvergenslatens och kommunikationsineffektivitet i peer-to-peer-nätverk.

Den första delen av avhandlingen fokuserar på decentralisering, konvergenshastighet och beräkningskostnad. Medan centraliserade ADMM-algoritmer lider av skalbarhetsproblem och flaskhalsar, förlitar sig existerande decentraliserade metoder antingen på omfattande beräkningar och kommunikation eller helt sekventiella operationer, vilket leder till långsammare konvergens. För att hantera dessa problem introducerar vi två snabbt konvergerande ADMM-algoritmer. Vår teoretiska analys bekräftar att dessa algoritmer behåller de klassiska konvergensegenskaperna hos centraliserad ADMM samtidigt som de bibehåller låg komplexitet på $O(1)$ per nod. Numeriska simuleringar visar vidare att våra algoritmer konvergerar betydligt snabbare än befintliga decentraliserade implementeringar.

Den andra delen av avhandlingen behandlar det klassiska problemet med eftersläntrare hos ADMM-algoritmer, där snabbare noder tvingas vänta på de långsammaste noderna på grund av det konventionella kravet på synkronisering i ADMM. Här introducerar vi tre algoritmer för att låta systemet förbli produktivt även under scenarier med en enda felpunkt eller extrem hårdvaruvarians. Den första algoritmen uppnår tolerans mot långsamma noder (eftersläntrare) genom att låta noderna fortsätta till nästa iteration även när en eller flera noder inte har tillhandahållit en uppdatering under en eller flera iterationer. Den andra algoritmen är en decentraliserad variant av den första algoritmen. Den tredje algoritmen utökar den andra algoritmen med ett tidsspårningsschema. Genom teoretiska analyser fastställer vi konvergensegenskaperna hos våra algoritmer och visar att våra decentraliserade algoritmer uppnår en beräkningskomplexitet på $O(1)$ för varje arbetsnod, medan vår centraliserade algoritm uppnår en beräkningskomplexitet på $O(N)$ för centralnoden och $O(1)$ per nod för de övriga noderna. Genom numeriska simuleringar med olika inställningar visar vi att våra algoritmer konvergerar betydligt snabbare än flera ADMM-algoritmer med väletablerade konvergensegenskaper.

Den sista delen av avhandlingen utökar dessa optimeringar till volatila system som kännetecknas av meddelandeförlust och dynamiska topologier. Här introducerar vi ytterligare två ADMM-algoritmer för dynamiska system, där nya noder kan ansluta sig mitt i processen samtidigt som systemet kan stöta på problem med eftersläntrare och meddelandeförlust. Dessa algoritmer är skapade för att uppnå snabb konvergens och flexibilitet genom att låta noder välja en stegstorlek som passar bäst för det egna systemet och genom att låta noderna gå vidare till nästa iteration även om inte alla noder har gjort en uppdatering för den aktuella iterationen. Ännu viktigare är att dessa algoritmer har ett identifieringssystem för att säkerställa konvergens trots meddelandeförlust. Algoritmerna upprätthåller även robusthet mot osäkerheter genom att ta bort behovet av att fördefiniera väntetiden eller det minsta antalet uppdateringar innan man går vidare till nästa iteration, genom ett tidsspårningsschema inspirerat av resultaten från den andra delen. Här har även en approximationsmekanism introducerats för att ge mer tid åt eftersläntrarna att utföra sina beräkningar medan de snabbare noderna går vidare till nästa iteration.

Sammanfattningsvis tillhandahåller denna avhandling accelererade algoritmer för snabb konvergens vid distribuerad optimering, med lösningar anpassade för såväl storskaliga som dynamiska nätverk.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology, 2026. , p. xi, 19
Series
TRITA-EECS-AVL ; 2026:67
Keywords [en]
ADMM, consensus optimization, distributed systems, asynchronous, decentralization, stragglers, communication
Keywords [sv]
ADMM, optimering, decentralisering, asynkrona operationer, dynamiska nätverk, kommunikation, algoritmer, linjär regression, distribuerad inlärning
National Category
Signal Processing Computer Vision and Learning Systems
Research subject
Computer Science
Identifiers
URN: urn:nbn:se:kth:diva-382160ISBN: 978-91-8106-660-9 (print)OAI: oai:DiVA.org:kth-382160DiVA, id: diva2:2061926
Public defence
2026-06-12, https://kth-se.zoom.us/j/64719519129, Q2, Malvinas väg 10, Stockholm, 13:30 (English)
Opponent
Supervisors
Note

QC 20260525

Available from: 2026-05-25 Created: 2026-05-23 Last updated: 2026-06-08Bibliographically approved
List of papers
1. Fast-Converging Decentralized ADMM for Consensus Optimization
Open this publication in new window or tab >>Fast-Converging Decentralized ADMM for Consensus Optimization
2024 (English)In: Proceedings - 2024 IEEE Conference on Artificial Intelligence, CAI 2024, Institute of Electrical and Electronics Engineers (IEEE) , 2024, p. 575-580Conference paper, Published paper (Refereed)
Abstract [en]

For its well-established convergence properties and applicability to various optimization problems, the alternating direction method of multipliers (ADMM) has been at the center of several research fields. When applied to distributed problems such as consensus optimization, ADMM is typically implemented in a centralized manner. Such implementations are, however, discouraged for e.g. their dependency on the location and capacity of the central node. While there are decentralized alternatives, these implementations are either computationally and communication-wise expensive or slow. This is because existing decentralized alternatives require all worker nodes to either replicate the work of synchronizing the outputs from all nodes or execute their tasks in sequence. To address this problem, we propose a fast-converging decentralized ADMM (FCD-ADMM) algorithm. Through theoretical analysis, we prove the convergence properties of FCD-ADMM and show that FCD-ADMM can converge faster than its centralized alternative without sacrificing accuracy. As shown in our numerical experiments, FCD-ADMM can converge to the same or better solution faster than several state-of-the-art alternatives.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2024
Keywords
Alternating Direction Method of Multipliers (ADMM), consensus optimization, convergence rate, decentralized optimization, distributed optimization
National Category
Control Engineering
Identifiers
urn:nbn:se:kth:diva-352376 (URN)10.1109/CAI59869.2024.00114 (DOI)001289387700104 ()2-s2.0-85201203781 (Scopus ID)
Conference
2nd IEEE Conference on Artificial Intelligence, CAI 2024, Singapore, Singapore, Jun 25 2024 - Jun 27 2024
Note

Part of ISBN [9798350354096]

QC 20240902

Available from: 2024-08-28 Created: 2024-08-28 Last updated: 2026-05-25Bibliographically approved
2. Fast-converging decentralized alternating direction method of multipliers for consensus optimization
Open this publication in new window or tab >>Fast-converging decentralized alternating direction method of multipliers for consensus optimization
2025 (English)In: EURASIP Journal on Advances in Signal Processing, ISSN 1687-6172, E-ISSN 1687-6180, Vol. 2025, no 1, article id 27Article in journal (Refereed) Published
Abstract [en]

For its well-established convergence properties, simplicity, and applicability to various optimization problems, the alternating direction method of multipliers (ADMM) has been at the center of several research fields. When applied to problems such as consensus optimization in distributed systems, ADMM is often implemented in a centralized manner, which may be challenging due to its dependency on the location and capacity of the node. While decentralized implementations have been proposed, these schemes require all worker nodes to either replicate the synchronization work of all nodes or execute their tasks in sequence, making them computationally and communication-wise expensive or slow. To tackle these challenges, we introduce two novel decentralized ADMM algorithms to enable decentralized learning without incurring redundant computations, excessive message transmissions, or extended convergence time. Through theoretical analysis, we have shown that our algorithms inherit the well-established convergence properties of the classical centralized ADMM while keeping the computational and communication complexity for each node at O(1). Moreover, we have identified specific conditions under which each of our algorithms is guaranteed to converge faster than the other as well as the classical centralized ADMM. Across several numerical simulations with various settings, our algorithms have converged several times faster than several state-of-the-art decentralized ADMM implementations.

Place, publisher, year, edition, pages
Springer Nature, 2025
Keywords
Alternating direction method of multipliers (ADMM), Decentralized optimization, Distributed machine learning, Consensus optimization
National Category
Control Engineering
Identifiers
urn:nbn:se:kth:diva-371931 (URN)10.1186/s13634-025-01225-8 (DOI)001529882700001 ()2-s2.0-105010726155 (Scopus ID)
Note

QC 20251022

Available from: 2025-10-22 Created: 2025-10-22 Last updated: 2026-05-25Bibliographically approved
3. Straggler-Resilient Asynchronous Decentralized ADMM for Consensus Optimization
Open this publication in new window or tab >>Straggler-Resilient Asynchronous Decentralized ADMM for Consensus Optimization
2024 (English)In: 2024 58TH ANNUAL CONFERENCE ON INFORMATION SCIENCES AND SYSTEMS, CISS, Institute of Electrical and Electronics Engineers (IEEE) , 2024Conference paper, Published paper (Refereed)
Abstract [en]

For its ability to combine the decomposability of dual ascent methods with superior convergence properties, the alternating direction method of multiplier (ADMM) has attracted substantial research interests and applications in various fields, e.g. distributed systems. Nevertheless, the problem of stragglers (slow-response nodes) and single-point-of-failures (a single node causing the failure of the entire system) remains. Therefore, we propose a novel asynchronous decentralized ADMM algorithm to address this. Through a series of simulations, our algorithm outperforms several state-of-the-art alternatives by converging multiple times faster in all settings. The results also show the resilience of our algorithm against stragglers and single points of failure as well as its ability to achieve superior convergence speed without sacrificing accuracy.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2024
Series
Annual Conference on Information Sciences and Systems, ISSN 2837-0163
Keywords
ADMM, consensus optimization, asynchronous, decentralization
National Category
Computer and Information Sciences
Identifiers
urn:nbn:se:kth:diva-350468 (URN)10.1109/CISS59072.2024.10480210 (DOI)001207282100053 ()2-s2.0-85190624601 (Scopus ID)
Conference
58th Annual Conference on Information Sciences and Systems (CISS), MAR 13-15, 2024, Princeton, NJ
Note

QC 20240715

Available from: 2024-07-15 Created: 2024-07-15 Last updated: 2026-05-25Bibliographically approved
4. Straggler-Resilient Asynchronous ADMM for Distributed Consensus Optimization
Open this publication in new window or tab >>Straggler-Resilient Asynchronous ADMM for Distributed Consensus Optimization
2025 (English)In: IEEE Transactions on Signal Processing, ISSN 1053-587X, E-ISSN 1941-0476, Vol. 73, p. 2496-2510Article in journal (Refereed) Published
Abstract [en]

For its simplicity, well-established convergence properties, and applicability to various optimization problems, the alternating direction method of multipliers (ADMM) has been widely used in several fields. However, when applied in distributed systems, the method may encounter the challenges of stragglers (nodes with significantly longer response time than others) and single points of failure (a single node causing the failure of the entire system). To address these problems, we propose three straggler-resilient ADMM algorithms. The first one is a centralized straggler-resilient ADMM algorithm achieving straggler-resilience by allowing the nodes to proceed to the next iteration even when one or more nodes have not provided an update for one or more iterations. The second one is an extension of the first one achieving single-point-of-failure resilience and fast convergence through decentralized, asynchronous, and concurrent operations. The third one is an extension of the second one to also achieve robustness against uncertainties with the help of a time-tracking scheme. Through theoretical analyses, we establish the convergence properties of our algorithms and show that our algorithms achieve a computational complexity of O(1) for each worker node - excluding the central node in the centralized algorithm, where the workload complexity is O(N). By numerical simulations with various settings, we show that our algorithms have converged significantly faster than several state-of-the-art ADMM algorithms with well-established convergence properties.

Place, publisher, year, edition, pages
Institute of Electrical and Electronics Engineers (IEEE), 2025
Keywords
ADMM, asynchronous, consensus optimization, decentralization, distributed systems, stragglers
National Category
Control Engineering
Identifiers
urn:nbn:se:kth:diva-368766 (URN)10.1109/TSP.2025.3579628 (DOI)001525462300002 ()2-s2.0-105008792161 (Scopus ID)
Note

QC 20250821

Available from: 2025-08-21 Created: 2025-08-21 Last updated: 2026-05-25Bibliographically approved
5. Robust Asynchronous Decentralized ADMM for Distributed Optimization in Dynamic Networks
Open this publication in new window or tab >>Robust Asynchronous Decentralized ADMM for Distributed Optimization in Dynamic Networks
(English)Manuscript (preprint) (Other academic)
Keywords
ADMM, consensus optimization, distributed systems, asynchronous, decentralization, stragglers, communication
National Category
Signal Processing
Research subject
Computer Science
Identifiers
urn:nbn:se:kth:diva-382105 (URN)
Note

QC 20260525

Available from: 2026-05-21 Created: 2026-05-21 Last updated: 2026-05-25Bibliographically approved

Open Access in DiVA

kappa(322 kB)87 downloads
File information
File name FULLTEXT03.pdfFile size 322 kBChecksum SHA-512
716204da3e80fe319bffb607cb2b61ab7d8ad024f7362618f60b81e1117c0310e20260ce371f6b1fe8f8fde041ca35692163a1878a3c29de8724c5fe2eb89c7f
Type fulltextMimetype application/pdf

Authority records

He, Jeannie

Search in DiVA

By author/editor
He, Jeannie
By organisation
Information Science and Engineering
Signal ProcessingComputer Vision and Learning Systems

Search outside of DiVA

GoogleGoogle Scholar
Total: 87 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

isbn
urn-nbn

Altmetric score

isbn
urn-nbn
Total: 614 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