MoReco: AI Model Recommendation and Optimization for B5G Networks

The deployment of Artificial Intelligence and Machine Learning (AIML) models in 5G networks has become increasingly critical for optimizing network performance, particularly in applications such as Self-Organizing Networks (SON) and Open Radio Access Networks (O-RAN). However, the dynamic nature of cloud-based network infrastructures presents significant challenges, as changes in underlying resources can adversely affect the inference time and prediction accuracy of these models. This paper introduces MoReco, a novel framework designed to predict and optimize the performance of AIML models in continuously changing deployment environments. MoReco features a “trade-off analyzer” that selects the most suitable ML algorithm and hyperparameters, ensuring that the trained model meets predefined thresholds for both inference time and accuracy. By maintaining a comprehensive record of previous iterations and dynamically tuning models in response to network changes, MoReco eliminates the need for repeated retraining and redeployment. The system also includes a predictive mechanism that estimates model performance without actual deployment, significantly improving the efficiency of the model deployment process. The proposed framework is evaluated within the context of 5G networks, demonstrating its potential to enhance reliability and operational efficiency in intelligent network management. The paper concludes by discussing future work aimed at expanding MoReco’s capabilities and exploring its application in other domains such as IoT and smart cities.