ADORN uses RL to adaptively handle drift in Open RAN AI/ML models

The paper introduces ADORN, a Q-learning-based adaptive retraining approach designed to mitigate performance degradation caused by dynamic traffic variations and drift in Open Radio Access Networks (O-RAN). By formulating the retraining decision as a Markov Decision Process, a Reinforcement Learning agent learns a policy that balances forecasting accuracy with retraining costs.

• The system incorporates a multi-expert Long Short-Term Memory (LSTM) ensemble to mitigate catastrophic forgetting and improve robustness across diverse traffic conditions.
• Experimental results demonstrate that ADORN effectively reduces retraining overhead compared to greedy and random baselines while maintaining system performance within predefined limits.

This approach addresses the high computational costs and potential Service Level Agreement violations associated with traditional retraining methods in O-RAN environments.