Performance Analysis and Optimization of Data-Driven AI Methods in Next Generation Networking

Abstract

Modern communication networks are experiencing a rapid transformation driven by the exponential growth of data traffic, the growing number of latency-sensitive applications, and the integration of heterogeneous computing and communication resources. Applications such as Large-Language Models (LLMs), real-time video analytics, extended reality (XR), industrial automation, and connected vehicles impose strict requirements on latency, reliability, scalability, and security. At the same time, computation and intelligence are increasingly moving from centralized cloud infrastructures toward the network edge, bringing new challenges for resource management, network control, and system security. Emerging communication technologies such as reconfigurable intelligent surfaces (RIS) and terahertz communication, together with the integration of AI and machine learning into network operations, further increase the complexity of future 6G networks. This thesis follows a holistic approach that combines mathematical modeling, data-driven optimization, reinforcement learning, security and resilience analysis, and intent-based automation, with the goal of enabling autonomous and adaptive control in edge to cloud infrastructures. First, the thesis develops mathematical models for analyzing and optimizing emerging networking technologies. These include optimization frameworks for the placement of deep neural network models in edge-computing environments, analytical models for frequency hopping optimization in LoRa networks, and data-driven channel estimation methods for terahertz communication systems. In addition, reliability-aware resource allocation and path selection models are introduced for RIS-assisted wireless networks. Second, the thesis investigates learning-based mechanisms for optimizing distributed edge networks. The proposed approaches include traffic prediction, topology change detection, and reinforcement learning methods for automated serverless scaling and machine learning model lifecycle management in edge-to-cloud systems. Third, the thesis advances traditional security mechanisms toward resilience and antifragility in future 6G networks. It introduces physical-layer vulnerability models for optical networks under jamming attacks, learning-based detection and mitigation schemes for WDM systems, and reliability analysis for RIS-assisted wireless links under hardware failures and environmental blockages. Finally, the thesis investigates Intent-Based Networking (IBN) as a key paradigm for automated network control and extends it toward learning-driven and agentic AI orchestration. It introduces mechanisms for intent translation, benchmarking ML workloads on heterogeneous hardware, intent drift detection, and intent security assurance. Thus, a multi-agent orchestration architecture is proposed in which specialized agents collaboratively interpret intents, monitor infrastructure, plan actions, and execute orchestration decisions. Finally, this thesis provides an integrated framework for modeling, optimizing, securing, and automating next-generation communication networks, contributing to the development of autonomous, intelligent, and secure 6G networking systems capable of supporting complex edge-to-cloud infrastructures.