Abstract
dc:descriptionThe rapid expansion of the cloud-edge computing continuum, fueled by the emergence of the Internet of Everything (IoE), has led to the development of highly heterogeneous, large-scale, and dynamic infrastructures. These environments are expected to support a wide spectrum of distributed applications—ranging from augmented and virtual reality and real-time multiplayer gaming to industrial IoT systems with strict latency constraints—operating across geographically dispersed and topologically complex ecosystems. Traditional infrastructure management approaches often rely on fragmented or isolated cognitive functions that fail to model or leverage the deep interdependencies inherent in such environments. This dissertation introduces a novel architectural paradigm, the Rhizomatic Cognitive Computing Continuum, which redefines how knowledge is organized and intelligent decision-making is manifested across the multilayered cloud-edge landscape. Central to this approach are two key concepts—rhizome and strata—adapted from the philosophical framework of Deleuze and Guattari. The rhizome serves as a dynamic, decentralized hypergraph that represents the entire continuum, where nodes capture heterogeneous informational entities—such as services, nodes, users, or contexts—and edges encode essential relationships, such as spatial, temporal, or causal links. Within this structure, strata emerge as modular, composable layers of cognitive functionality. Each stratum encapsulates a distinct cognitive task and reflects its scope of operation (e.g., local or global), forming an evolving cognitive stack. A key contribution of this dissertation lies in the design and implementation of deep learning models based on Graph Neural Networks (GNNs), each tailored to a specific cognitive stratum and operating on the shared hypergraph. For instance, the mobility prediction layer leverages the WEST GCN-LSTM architecture, integrating region- and population-sensitive policies to accurately forecast user movements. The local proactive autoscaling layer uses an encoder-decoder GNN to predict resource usage across time horizons, capturing local dependencies for efficient real-time response. The global autoscaling layer introduces GraphOpticon, a GNN system that aggregates data across clusters to optimize scaling actions globally. The service image placement layer utilizes GNOSIS, a hybrid GNN and reinforcement learning model that formulates deployment as a minimum vertex cover problem, reducing resource usage and latency. Finally, the SLA violation prediction layer employs composite GNN architectures that synthesize temporal, topological, and semantic information to enable accurate early warnings of service-level breaches. In addition, this work proposes a robust federated mechanism for managing cognitive strata across clusters, using the Cluster API and Liqo to orchestrate and interconnect multiple computing domains, ensuring continuous and consistent operation even in highly dynamic or fragmented environments. By decoupling cognitive functions into modular strata and grounding them within a semantically rich, shared rhizomatic space, the Rhizomatic Cognitive Computing Continuum addresses the limitations of conventional approaches. It enables scalable, context-aware, and synergistic decision-making aligned with the requirements of next-generation cyber-physical systems. This dissertation offers not only a conceptual reframing of the problem space but also practical implementations of the strata as deployable and interoperable cognitive agents, laying the foundation for a more intelligent and self-adaptive computing continuum.
Degree
thesis:*- Grantor dc:publisher
- Harokopio University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Theodoropoulos, Theodoros
- Θεοδωρόπουλος, Θεόδωρος
Subjects
dc:subject × 10- Μηχανική μάθηση
- Κατανεμημένα συστήματα λογισμικού
- Machine learning
- Distributed computing
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/59662