Glasgow Researchers Use Machine Learning to Build Network Digital Twin

Why It Matters

Accelerating network testing by orders of magnitude reduces operational costs and enables rapid rollout of resilient digital infrastructure, benefiting both telecom and emerging transport‑network applications.

Key Takeaways

• AutoML‑generated digital twins test networks 25,000× faster than simulators
• Testing of 12‑node and 37‑node topologies completed in under 5 seconds
• Approach requires minimal ML expertise, opening network testing to non‑experts
• Potential to extend digital twin factory concept to transport network optimization

Pulse Analysis

Network testing has long been a bottleneck for operators, relying on detailed simulators that can take days to model complex traffic patterns. Traditional tools demand deep domain expertise and extensive computational resources, limiting their use to large enterprises. Digital twins—virtual replicas of physical systems—offer a way to emulate behavior in real time, but building them manually is labor‑intensive. By integrating AutoML, researchers can automate twin creation, dramatically shortening the development cycle and democratizing access to sophisticated testing capabilities.

The Glasgow team’s breakthrough demonstrates that an AutoML‑driven twin can evaluate a 12‑node and a 37‑node network across six traffic scenarios in just 4.78 seconds, a speedup of roughly 25,000 times over a conventional simulator that required 33 hours. This performance leap stems from the twin’s ability to learn network dynamics from data rather than exhaustively simulating every packet. The approach maintains high accuracy while slashing compute costs, making it attractive for telecom carriers facing exploding data volumes and for enterprises seeking rapid validation of new protocols or security patches.

Beyond telecommunications, the research aligns with the broader TransiT initiative to create a “digital twin factory” for transport infrastructure. As vehicle connectivity and smart‑city sensors generate massive data streams, similar testing challenges arise. Automated digital twins could enable planners to model traffic flow, assess decarbonisation pathways, and optimize routing without costly field trials. The scalability and low‑skill barrier of the Glasgow method position it as a catalyst for cross‑industry adoption, promising faster innovation cycles and more resilient networked systems.