Abstract:Beyond fifth-generation (B5G) networks aim to support high data rates, low-latency applications, and massive machine communications. Artificial Intelligence/Machine Learning (AI/ML) can help to improve B5G network performance and efficiency. However, dynamic service demands of B5G use cases cause AI/ML model performance degradation, resulting in Service Level Agreements (SLA) violations, over- or under-provisioning of resources, etc. Retraining is essential to address the performance degradation of the AI/ML models. Existing threshold and periodic retraining approaches have potential disadvantages, such as SLA violations and inefficient resource utilization for setting a threshold parameter in a dynamic environment. This paper proposes a novel approach that predicts when to retrain AI/ML models using Generative Artificial Intelligence. The proposed predictive approach is evaluated for a Quality of Service Prediction use case on the Open Radio Access Network (O-RAN) Software Community platform and compared to the predictive approach based on the classifier and a threshold approach. Also, a realtime dataset from the Colosseum testbed is considered to evaluate Network Slicing (NS) use case with the proposed predictive approach. The results show that the proposed predictive approach outperforms both the classifier-based predictive and threshold approaches.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of performance degradation of artificial intelligence/machine learning (AI/ML) models in the fifth - generation mobile communication (5G) and beyond (Beyond 5G, B5G) networks. Specifically: 1. **Performance degradation caused by dynamic service requirements**: The dynamic service requirements in B5G networks can lead to a decline in the performance of AI/ML models, resulting in issues such as service - level agreement (SLA) violations, over - or under - allocation of resources. 2. **Limitations of existing retraining methods**: - **Threshold retraining method**: Trigger retraining by setting thresholds for performance indicators (such as accuracy, precision, recall, etc.). However, in a dynamic environment, it is difficult to set appropriate thresholds using this method, which may lead to frequent retraining or poor performance. - **Periodic retraining method**: Retrain the model regularly regardless of changes in user traffic or model performance. This may lead to waste of resources or failure to respond to performance changes in a timely manner. - **Classifier - based prediction method**: Use an unsupervised classifier (such as local outlier factor LOF) to predict when retraining is required. However, this method is not effective in dealing with non - continuous changes or high - variance user traffic. 3. **The proposed new method**: To overcome the limitations of the above methods, the paper proposes a new method based on generative artificial intelligence (Generative AI, Gen - AI), using variational auto - encoder (VAE) and generative adversarial network (GAN) to predict when AI/ML models need to be retrained. By generating synthetic data similar to the observed data and using the Kolmogorov - Smirnov (KS) test to detect changes in user traffic, it is determined whether the model needs to be retrained. ### Main contributions - **Proposed a new Gen - AI - based prediction method**: Use VAE and GAN to generate synthetic data close to the observed data, and use the KS test to predict when AI/ML models need to be retrained. - **Evaluated the effectiveness of the new method**: Evaluated on the QoS prediction use case on the O - RAN software community platform and the real - time data set of the Colosseum testbed, and compared with the classifier - based prediction method and the threshold method. - **Discussed the challenges and complexity of applying Gen - AI in B5G networks**: Including applications in generating synthetic data, resource optimization, network planning, and QoS management. Through these improvements, the paper aims to improve the performance stability of AI/ML models in B5G networks, reduce SLA violations, and optimize resource utilization.

Generative-AI for AI/ML Model Adaptive Retraining in Beyond 5G Networks

Design of an Iterative Method for Dynamic Resource Management in 5G Networks with IoT Integration Operations

Toward Scalable and Efficient Hierarchical Deep Reinforcement Learning for 5G RAN Slicing

Deep Learning in Beyond 5G Networks with Image-based Time-Series Representation

ADAPTIVE6G: Adaptive Resource Management for Network Slicing Architectures in Current 5G and Future 6G Systems

Intelligent Propagation Model Method for RSRP Prediction Based on Machine Learning

Interference-Aware Intelligent Scheduling for Virtualized Private 5G Networks

Deep Learning for B5G Open Radio Access Network: Evolution, Survey, Case Studies, and Challenges

ECO6G: Energy and Cost Analysis for Network Slicing Deployment in Beyond 5G Networks

X-GRL: An Empirical Assessment of Explainable GNN-DRL in B5G/6G Networks

Artificial Intelligence Enabled Wireless Networking for 5G and Beyond: Recent Advances and Future Challenges

Deep reinforcement learning‐aided RAN slicing enforcement supporting latency sensitive services in B5G networks

Reasoning AI Performance Degradation in 6G Networks with Large Language Models

AI-Empowered Propagation Prediction and Optimization for Reconfigurable Wireless Networks

Future Vision on Artificial Intelligence Assisted Green Energy Efficiency Network

AI-Chain: Blockchain Energized Edge Intelligence for Beyond 5G Networks

Machine Learning at the Edge: A Data-Driven Architecture With Applications to 5G Cellular Networks

A Multi-Agent Deep Reinforcement Learning Approach for RAN Resource Allocation in O-RAN

Machine Learning in Beyond 5G/6G Networks—State-of-the-Art and Future Trends

Empirical performance analysis and ML-based modeling of 5G non-standalone networks

Intelligent Traffic Steering in Beyond 5G Open RAN based on LSTM Traffic Prediction