Omar Houidi

• Designed an end-to-end network slicing framework enhanced by Transformer-based DRL agents, improving dynamic adaptability and service differentiation in future 5G/6G environments. • Proposed energy-aware VNF-FG placement strategies leveraging attention-based deep reinforcement learning and transformer architectures to optimize resource utilization and reduce power consumption in NFV infrastructures. • Developed monitoring-aware network slicing frameworks for 5G Core, integrating observability metrics into slice orchestration decisions for improved performance and SLA compliance. • Conducted research on routing and spectrum assignment (RSA) in elastic optical networks (EON) using multi-agent reinforcement learning and graph-based models (MAGC-RSA), enabling intelligent spectrum allocation in core optical infrastructure.

Project: NS-3 Extension for Large-scale 5G SA and Open RAN Simulation • Designed and developed NS-3 extensions to simulate large-scale 5G Standalone (SA) networks incorporating O-RAN architectures. • Extended the 5G-LENA simulator to support key features of 3GPP Release 17, including control/user plane integration, beamforming, and slicing. • Implemented enhancements in NS-3 for 5G gNB (SA mode): control plane development, GTP tunneling, initial UE attachment, handovers, and UE release. • Integrated Open RAN (O-RAN) E2 interfaces to simulate disaggregated gNB with scalable control via service models. • Combined user/control plane elements from UERANSIM and Open5GS to support flexible and hybrid simulation setups. • Enabled support for AI/ML-driven experimentation, dataset generation, and RAN control algorithm evaluation in O-RAN environments. • Target use cases: XR/VR traffic simulation, slicing validation, performance benchmarking of intelligent RIC behaviors.

Project: Distributed DRL-based Multi-Agent System (MAS) for Intelligent Routing, Load Balancing, and Smart Queuing. • Designed and developed a distributed AI-based routing system combining multi-agent cooperation and Deep Reinforcement Learning (DRL). • Proposed a Multi-Agent System (MAS) architecture to address scalability, reactivity, and fault tolerance limitations of centralized SDN/IGP solutions. • Implemented actor-critic reinforcement learning algorithms for load balancing and smart queuing in dynamic network environments. • Integrated Graph Neural Networks (GNNs) and Attention mechanisms for enhanced agent cooperation and decision-making. • Novel use of Graph Convolutional Reinforcement Learning. • Hybrid agent architecture with centralized training, distributed execution. • Use of attention mechanisms to reduce agent communication overhead. • Validated system performance against IGP and centralized SDN benchmarks: - SLA violations reduced by 20%. - Achieved 10× faster convergence when faced with network changes, compared to the SDN baseline. - Scale-tested for up to 100+ nodes with no routing loops.

PhD Thesis: Algorithms for Virtual Network Functions Chaining. • Designed and implemented exact (ILP) and heuristic algorithms for efficient placement and chaining of virtualized network services (VNF-FGs). • Addressed dynamic and scalable orchestration in NFV/SDN environments with a focus on service elasticity and real-time adaptation. • Explored VNF graph extension, scaling, and reinforcement learning for smart service deployment. • SDN, NFV, Service Function Placement, Chaining and Orchestration. • Smart and dynamic deployment of VNF-FGs. • Strong focus on Service Function Placement, Elastic SFCs, and network slicing in next-generation architectures (5G/Edge). • Technologies: NFV, SDN, Service Function Chaining (SFC), ILP modeling, Steiner Trees, Q-Learning.