Abstract
dc:descriptionIn today's world, information gathering, communication and processing constitutes the pillar of modern infrastructure that aims in creating a better environment and quality of life for end-users. The Next Generation Internet of Things (NG-IoT) is a network of intelligent devices that produce, communicate and exchange this information providing more accurate and personalised services to the end-user. NG-IoT builds on top of conventional IoT by bringing enhanced data production capabilities analogous to the ability to host on-board intelligence for in-depth processing and decision support. As modern infrastructure becomes more data-intensive and complex, new tools are needed to support the ingestion and processing of the generated information.Artificial Intelligence (AI), along with its fields Machine Learning (ML) and Deep Learning (DL), is crucial for advancing NG-IoT practices as it provides the baseline for achieving optimization of operations, services, and resources through advanced analytical capabilities. The research community has expressed an increasing interest in this combination, as it has the premise to create a paradigm shift in modern infrastructure operations. Alas, this powerful tool is not without its challenges. Challenges in the AI field include how to train AI models optimally while leveraging large-scale NG-IoT data along with ways to optimize it, addressing computation, resources, and security/privacy concerns.Federated Learning (FL), a novel AI model optimisation approach for training models on decentralised corpora of remote devices, has emerged as the solution to tackle the obstacles that arise from integrating AI into the edge ecosystem. Federated Learning also addresses the security and privacy risks of pooling data from varied and dispersed sources in conventional Machine Learning methods, while reducing communication/computation overheads, ensuring remote data integrity, confidentiality, and privacy. Unfortunately, FL like any other novel technology, brings its own preconditions and challenges, intrinsically related to the complexity and uncertainty of the decentralised environment, the variations of decentralised data (non-independent and identically distributed - non-IID), scheduling, and, even social fairness between interacting NG-IoT entities. This thesis aims to design, develop and implement Federated Learning techniques in the decentralised NG-IoT ecosystem. In particular, the main work of this thesis is focused on designing, applying, and validating new and novel FL techniques in NG-IoT, leveraging state-of-the-art methodologies, while trying to find solutions for emerging applications of FL. To this end, this work focuses on three main pillars of applying FL to NG-IoT applications, namely, i) Architectures and Methodologies for Privacy-Preserving FL in NG-IoT, ii) Optimization of FL Models under Privacy Constraints, and iii) Ensuring Fairness and Equity in Federated Learning.
Degree
thesis:*- Grantor dc:publisher
- University of Western Macedonia
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Siniosoglou, Ilias
- Σινιόσογλου, Ηλίας
Subjects
dc:subject × 12- Τεχντητή νοημοσύνη
- Ομοσπονδιακή μάθηση
- Διαδίκτυο επόμενης γενιάς
- Artificial intelligence
- Federated learning
- Next generation networks
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Μηχανική και συστήματα επικοινωνιών, Τηλεπικοινωνίες
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Communication engineering and systems, Telecommunications
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60433