Demo: A Drift-Handling Automation in AI/ML Framework Integrated O-RAN

Beyond 5G (B5G) networks increasingly leverage open and disag-
gregated Radio Access Networks (O-RAN) to enable high-speed,
low-latency, and flexible services across diverse use cases. How-
ever, the dynamic nature of user traffic and network conditions
causes Artificial Intelligence (AI)/ Machine Learning (ML) models
in RAN Intelligent Controllers (RICs) to experience performance
drift – leading to inefficient radio resource allocation and potential
service level agreements (SLA) violations. This work demonstrates
a real-time drift detection, analysis, and adaptation by leveraging an
AI/ML framework integrated O-RAN. This is achieved with real-
time data collection from open interfaces which would be used by
intelligent controllers (i.e., Non-RT RIC) for training and inference
purposes of AI/ML models. The proposed drift-handling automa-
tion monitors network performance metrics to identify model drift
with minimal data and computation overheads and then adapts
to current network conditions by either retraining the model or
replacing it with an available pre-trained model. This approach
improves reliability, agility, and automation in multi-vendor B5G
deployments by means of drift detection and resolution.