Back to search

National and Kapodistrian University of Athens

Σύστημα επίγνωσης πλαισίου και διαφύλαξης της ιδιωτικότητας για κατανεμημένη και ενεργειακά αποδοτική τεχνητή νοημοσύνη σε δίκτυα νέας γενιάς

Abstract

dc:description

The advent of fifth-generation networks represents a significant milestone in the evolution wireless communications. Since 2018, discussions have emerged regarding Beyond 5G and 6G networks, which are expected to introduce a set of novel challenges for the next generation of cellular networks. The vast number of heterogeneous devices contending for network resources will produce massive amounts of information that must be managed for efficient resource planning and optimized network performance under varying network conditions. Reliable network communication is crucial, as the increased density of Beyond 5G and 6G networks could negatively impact Key Performance Indicators, leading to network failures. As a result, autonomous systems must integrate proactive mechanisms to prevent performance degradation. Energy efficiency is also vital as network density and connected devices increase, raising energy demands. The era of Network Intelligentization and Connected Intelligence will introduce additional energy challenges, requiring trade-offs between energy efficiency and effective network operation. Privacy-aware distributed networks will be essential for protecting sensitive information in B5G and 6G networks, necessitating new distributed architectures where devices perform local computations collaboratively. However, these limitations may complicate the implementation of such mechanisms in large-scale networks with stringent performance and latency requirements. In parallel, the use of Artificial Intelligence has been recognized by industry and academia as a key enabler for Beyond 5G and 6G networks. However, Artificial Intelligence introduces new challenges, including communication overhead and energy requirements, privacy constraints in data exchanges, constantly changing network channel conditions and fluctuations in traffic congestion, and device heterogeneity, in terms of data types, data distribution, and computational capabilities. To this end, the core principle of this thesis is the introduction of a novel, holistic, comprehensive system that incorporates cutting-edge practices and innovative mechanisms and techniques, enabling a systematic approach to addressing the significant challenges of B5G and 6G networks. More specifically, this thesis proposes a novel context- and privacy-aware system for distributed and energy efficient artificial intelligence in next generation networks. It consists of five main contributions, with four of them presented as individual frameworks while the fifth one is related to a conceptual deployment approach in future wireless communications. The four frameworks are designated as PRIMATE (Profiling Mechanism Based on Context), NINTENDO (Network Traffic Anomaly Prediction), SMART (Safe Deep Reinforcement Learning for Energy Efficient Federated Learning) and DISTINQT (Distributed Privacy Aware Learning for QoS Prediction). The first major contribution of this thesis in PRIMATE. PRIMATE describes an Artificial Intelligence-driven context-aware profiling framework capable of monitoring, aggregating, analyzing and combining heterogeneous knowledge from various sources. All this knowledge is used to extract context. This context is dynamically projected into a new latent representation, capturing different behavioral aspects of the network, designated as behavioral profiles. Theses behavioral profiles can be used to efficiently differentiate, manage, and predict the resources allocated to heterogeneous types of network entities, and can be ultimately exploited by network domain experts and administrators towards efficient and proactive resource planning. The second major contribution of this thesis is NINTENDO. NINTENDO describes an Artificial Intelligence-driven framework capable of proactively detecting network traffic anomalies. NINTENDO's contribution is twofold. It describes a mechanism aimed at identifying different network traffic behaviors in an automated manner, while it proposes a mechanism for predicting network traffic behaviors for the next time interval spanning several seconds. This approach provides the network with the ability to take adaptation actions in a proactive manner, in order to potentially prevent an impending network failure. NINTENDO acts as a practical implementation of PRIMATE, demonstrating its capabilities and extends it by introducing a profiling analysis module capable of quantitatively analyzing the characteristics of each behavioral profile. The third major contribution of this thesis is SMART. SMART describes an Artificial Intelligence-driven framework that aims to orchestrate the computational and communication resources of the devices involved in a federated learning process. The orchestration of resources is achieved through the implementation of a reinforcement learning algorithm, which enables real-time decision-making and addresses the heterogeneity inherent in both the devices and the data, which is subject to change over time. This framework builds upon the PRIMATE framework and serves as a proof of concept regarding the assertions made by PRIMATE regarding efficient resource planning while also exploiting the profiling analysis module proposed by the NINTENDO framework. The fourth major contributions of this thesis is DISTINQT. DISTINQT introduces a multi-headed input privacy-aware distributed learning framework for QoS prediction. DISTINQT is capable of supporting multiple heterogeneous nodes, in terms of data types and model architectures, by sharing computations across them. This enables the incorporation of diverse knowledge into a sole learning process that enhance the robustness and generalization capabilities of the final QoS prediction model. DISTINQT also contributes to data privacy preservation by encoding any raw input data into highly complex, compressed, and irreversible latent representations. Furthermore, this framework builds upon the PRIMATE framework to distribute the core forecasting mechanisms of PRIMATE across multiple computing nodes, facilitating computation offloading and privacy preservation. The fifth and final major contribution is a conceptual deployment approach for the presented novel comprehensive system, using the latest architectural specifications defined by 3rd Generation Partnership Association as a reference point and extending it using the latest 5G Advanced technologies and foreseen 6G technologies.

Degree

thesis:*
Grantor dc:publisher
National and Kapodistrian University of Athens
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Koursioumpas, Nikolaos
  • Κουρσιουμπάς, Νικόλαος

Subjects

dc:subject × 16

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/59658

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Koursioumpas, Nikolaos; Κουρσιουμπάς, Νικόλαος. Σύστημα επίγνωσης πλαισίου και διαφύλαξης της ιδιωτικότητας για κατανεμημένη και ενεργειακά αποδοτική τεχνητή νοημοσύνη σε δίκτυα νέας γενιάς. National and Kapodistrian University of Athens, 2025. http://hdl.handle.net/10442/hedi/59658