In-Network Computing and Net4AI Paradigms For Future 6G Programmable Networks

Abstract

The evolution toward 6G networks presents unprecedented opportunities for integrating Artificial Intelligence (AI) within network architectures, transforming traditional communication systems into intelligent, adaptable infrastructures. This thesis explores two key research directions that advance this vision: in-network security and network support to distributed AI. The first line of research addresses the need for secure-by-design principles by proposing a distributed, in-network security framework. Leveraging programmable networking devices, this framework embeds anomaly detection models directly within the data plane, creating a pervasive security fabric capable of identifying and mitigating malicious activities in real-time. This approach emphasizes the critical role of distributed intelligence in achieving robust, resilient networks. The second line of research utilizes In-Network Computing (INC) and Split-AI techniques to distribute AI-relevant computational tasks across various network elements. By dynamically offloading parts of complex AI models, e.g. Neural Network (NN), to computation-enabled user plane entities, this approach reduces latency, conserves energy, and minimizes processing demands on User Equipment (UE). Through the development of an Intelligent User Plane (IUP), this thesis demonstrates how next-generation networks can natively support distributed AI applications, enabling real-time, resource-efficient processing closer to end-users. Together, these contributions underscore the potential of AI-enhanced 6G networks to deliver secure, scalable, and intelligent infrastructure. By addressing critical requirements for both security and AI support, this work provides a comprehensive foundation for the design and deployment of resilient, AI-native network systems in the 6G era.