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
Multiobjective and Constrained Reinforcement Learning for IoT
Stockholm University, Faculty of Social Sciences, Department of Computer and Systems Sciences.
Stockholm University, Faculty of Social Sciences, Department of Computer and Systems Sciences.ORCID iD: 0000-0002-6617-8683
Number of Authors: 22024 (English)In: Learning techniques for Internet of Things / [ed] Praveen Kumar Donta; Abhishek Hazra; Lauri Lovén, Springer , 2024, p. 153-170Chapter in book (Refereed)
Abstract [en]

IoT networks of the future will be characterized by autonomous decision-making by individual devices. Decision-making is done with the purpose of optimizing certain objectives. A multitude of mathematically oriented algorithms exist for solving optimization problems. However, optimization in IoT networks is challenging due to a number of uncertainties, complex network topologies, and rapid changes in the environment. This makes the data-driven and machine learning (ML) approaches more suitable for effectively handling IoT environments’ dynamic and intricate nature. However, supervised and unsupervised ML approaches depend on training data, which is not always available before training. In recent years, reinforcement learning (RL) has attracted considerable attention for solving optimization problems in IoT. This is because RL has the distinguishing feature of learning with experience while interacting with the environment without training data. A central challenge in decision-making in IoT networks is that most optimization problems consist of co-optimizing multiple conflicting objectives. With the development of multi-objective RL (MORL) approaches over the last two decades, there is great potential for utilizing them for future IoT networks. Most recently developed MORL approaches have not been applied in the IoT domain. In this chapter, we will discuss the need for efficient multi-objective optimization in IoT, the fundamentals of using RL for decision-making in IoT, an overview of existing MORL approaches, and, finally, the future scope and challenges associated with utilizing MORL for IoT.

Place, publisher, year, edition, pages
Springer , 2024. p. 153-170
Keywords [en]
Multiobjective Reinforcement Learning, Optimization, Internet of Things
National Category
Computer Sciences
Research subject
Computer and Systems Sciences
Identifiers
URN: urn:nbn:se:su:diva-227532DOI: 10.1007/978-3-031-50514-0_8Scopus ID: 2-s2.0-85206043207ISBN: 978-3-031-50513-3 (print)ISBN: 978-3-031-50514-0 (electronic)OAI: oai:DiVA.org:su-227532DiVA, id: diva2:1845290
Available from: 2024-03-18 Created: 2024-03-18 Last updated: 2026-04-13Bibliographically approved
In thesis
1. AI-Driven Multi-objective Decision-Making With Applications to IoT
Open this publication in new window or tab >>AI-Driven Multi-objective Decision-Making With Applications to IoT
2026 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Artificial intelligence (AI) plays an increasingly central role in enabling autonomous decision-making in complex, uncertain environments. Many modern systems must optimise multiple, often conflicting objectives while operating under dynamic resource constraints and incomplete knowledge of system dynamics. Classical approaches such as dynamic programming, constrained stochastic optimisation, and static multi-objective scalarisation provide principled solutions when accurate models are available. However, in distributed and stochastic environments such as the Internet of Things (IoT), system dynamics are often unknown, non-stationary, and resource-limited, making purely model-based methods difficult to apply.

Reinforcement learning (RL) and online learning offer an alternative by enabling policy adaptation through interaction rather than relying on explicit system models. Within multi-objective settings, existing approaches often assume fixed scalarisation weights or externally specified preferences and typically focus on learning policies for given trade-offs. In dynamic IoT systems, however, both resource constraints and preference parameters may vary over time, requiring algorithms that can adapt efficiently without repeated retraining or centralised coordination.

This thesis investigates how AI-based methods, with a primary focus on reinforcement learning and complementary distributed learning techniques, can support adaptive multi-objective decision-making under communication constraints, explicit resource limitations, and dynamically changing trade-offs. The research is organised around three themes. First, communication-efficient distributed learning methods are developed to balance model accuracy and communication cost infederated learning through adaptive sparsification. Second, constrained bandit and reinforcement learning formulations are proposed to incorporate explicit and time-varying resource constraints while maintaining theoretical performance guarantees. Third, multi-objective reinforcement learning methods are designed to adapt routing decisions indistributed IoT systems under dynamically changing energy–reliabilitytrade-offs without retraining.

Overall, the thesis demonstrates that integrating communication awareness, constraint handling, and preference adaptation directly into learning algorithms is essential for reliable AI-based decision-makingin IoT environments. The results provide both algorithmic advances and a conceptual framework for designing autonomous systems that operate robustly under dynamic objectives and limited resources.

Abstract [sv]

Artificiell intelligens (AI) spelar en allt viktigare roll för att möjliggöra autonomt beslutsfattande i komplexa och osäkra miljöer. Många moderna system måste optimera flera, ofta motstridiga mål samtidigt som de verkar under dynamiska resursbegränsningar och med ofullständig kunskap om systemets dynamik. Klassiska metoder såsom dynamisk programmering, begränsad stokastisk optimering och statisk multiobjektiv skalärisering erbjuder principiella lösningar när exakta modeller finns tillgängliga. I distribuerade och stokastiska miljöer, såsom Internet of Things (IoT), är dock systemdynamiken ofta okänd, icke-stationär och resursbegränsad, vilket gör rent modellbaserade metoder svåra att tillämpa.

Förstärkningsinlärning (reinforcement learning, RL) och onlineinlärning erbjuder ett alternativ genom att möjliggöra policyanpassning genom interaktionsnarare än genom explicita systemmodeller. Inom multiobjektiva problemantar befintliga metoder ofta fasta skaläriseringsvikter eller externt specificerade preferenser och fokuserar främst på att lära policyer för givna avvägningar.I dynamiska IoT-system kan dock både resursbegränsningar och preferensparametrarförändras över tid, vilket kräver algoritmer som kan anpassa sig effektivtutan upprepad ominlärning eller centraliserad styrning.

Denna avhandling undersöker hur AI-baserade metoder, med särskilt fokus på förstärkningsinlärning och kompletterande distribuerade inlärningsmetoder,kan stödja adaptivt multiobjektivt beslutsfattande under kommunikationsbegränsningar,explicita resurskrav och dynamiskt föränderliga avvägningar.Forskningen är organiserad kring tre huvudteman. För det första utveckla kommunikationseffektiva distribuerade inlärningsmetoder som balanserar modellnoggrannhet och kommunikationskostnad i federerad inlärning genom adaptiv sparsifiering. För det andra föreslås begränsade bandit- och förstärkningsinlärningsformuleringar för att hantera explicita och tidsvarierande resursbegränsningar med teoretiska prestandagarantier. För det tredje utvecklasmultiobjektiv förstärkningsinlärning för att möjliggöra adaptiv routingi distribuerade IoT-system under dynamiskt föränderliga avvägningar mellanenergiförbrukning och tillförlitlighet, utan behov av ominlärning.

Sammanfattningsvis visar avhandlingen att det är avgörande att explicit integrera kommunikationsmedvetenhet, begränsningshantering och preferensanpassning i inlärningsalgoritmer för att uppnå tillförlitligt AI-baserat beslutsfattande i IoT-miljöer. Resultaten bidrar med både algoritmiska framsteg och konceptuell grund för att utforma autonoma system som kan verka robusta under dynamiska mål och begränsade resurser.

Place, publisher, year, edition, pages
Stockholm: Department of Computer and Systems Sciences, Stockholm University, 2026. p. 116
Series
Report Series / Department of Computer & Systems Sciences, ISSN 1101-8526 ; 26-005
Keywords
Artificial Intelligence, Multi-objective, Internet of Things, Federated Learning, Reinforcement Learning
National Category
Communication Systems Computer Systems Telecommunications Other Electrical Engineering, Electronic Engineering, Information Engineering
Research subject
Computer and Systems Sciences
Identifiers
urn:nbn:se:su:diva-254165 (URN)978-91-8107-602-8 (ISBN)978-91-8107-603-5 (ISBN)
Public defence
2026-05-29, Small Auditorium NOD (Lilla hörsalen), plan 2, Borgarfjordsgatan 12, Kista, 13:00 (English)
Opponent
Supervisors
Available from: 2026-05-06 Created: 2026-04-13 Last updated: 2026-04-28Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Vaishnav, ShubhamMagnússon, Sindri

Search in DiVA

By author/editor
Vaishnav, ShubhamMagnússon, Sindri
By organisation
Department of Computer and Systems Sciences
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 148 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