Abstract The growing complexity of modern networks drives the need for Network Digital Twins (NDT)—virtual replicas enabling real-time monitoring, simulation, and predictive management. While promising for automated 5G and beyond-5G management, operational NDTs face challenges in continuous telemetry ingestion and network behavior modeling. This work presents a cloud-native monitoring and prediction framework built on a modular, containerized AWS architecture supporting Kubernetes and SDN. It fuses real-time telemetry with machine learning to forecast critical performance indicators (CPU, memory stress) and system failures (Alert Severity Level), enabling safe evaluation of configuration changes. Experimental validation under stressed workloads demonstrates accurate degradation predictions and proactive closed-loop control. Aligned with IRTF NMRG reference architectures, this provides actionable directions toward AI-enhanced autonomous network management and discusses carrier-grade deployment and non-terrestrial network integration for next-generation systems.

Abstract High-fidelity digital twin networks require virtual models that capture tail latency and jitter, not just average performance. Existing ML‑based simulators, trained with mean squared error, are “tail‑blind” and break the fidelity needed for dependable digital twins of latency‑sensitive systems. We introduce Accurate Neural Architecture Search, which automatically discovers neural architectures that faithfully model the full delay distribution. It corrects weight‑sharing NAS evaluation bias via a similarity‑constrained search, uses a hybrid search space to represent complex traffic, and employs a Wasserstein loss to optimize the entire distribution. The discovered architecture reduces validation loss by 25.8% over existing methods and cuts normalized Wasserstein distance by up to 69.8%, delivering both strong average‑case and tail performance. It thus provides an automated, practical methodology for building the precise virtual replicas that underpin simplified, resilient, and fully automated network operations in digital twin systems.

Abstract The Network Digital Twin (NDT), a real-time virtual representation of a telecommunications network, is emerging as a foundational enabler for 6G, allowing service providers to replicate, predict, and optimize network behavior without impacting the actual Network. This presentation covers the NDT technical architecture, key use cases, and how different AI methods can enable an efficient and accurate network digital twin. We discuss how NDT with predictive and prescriptive AI enables use cases such as autonomous network operations, network security and proactive SLA assurance. We also highlight how hybrid approaches combining physics-based simulation and data-driven models address the challenge of replicating real network topology, propagation environment, and dynamic network behavior.

Abstract Future networks face growing complexity from heterogeneous devices, dynamic services, and diverse requirements such as latency, reliability, energy efficiency, and security. This talk presents an intent-driven digital twin network framework for autonomous management. Large language models and semantic parsing translate natural-language intents into policy representations, while digital twins validate, optimize, and refine policies before deployment. An IoT-oriented case study demonstrates intent translation for devices, protocols, and security configurations. The talk also discusses challenges in trustworthy policy generation, real-time twin synchronization, and scalable closed-loop autonomous networking.

Abstract Digital twin technology serves as a key enabler for the intelligent upgrading of autonomous networks. It builds high-fidelity virtual replicas of physical network infrastructures to mirror real-time network states and operational features. Combined with artificial intelligence and closed-loop simulation, digital twins empower 5G-A and 6G radio access networks with intelligent sensing, predictive analysis, dynamic optimization and autonomous decision-making capabilities. This approach efficiently tackles the complexity and variability of modern wireless networks, streamlines network planning, operation and maintenance workflows, and boosts network efficiency and resource utilization, which is critical for achieving high-level autonomy in future mobile networks.