Adapting MLOps for Diverse In-Network Intelligence in 6G Era: Challenges and Solutions

Seamless integration of artificial intelligence (AI) and machine learning (ML) techniques with wireless systems is a crucial step for 6G AInization. However, such integration faces challenges in terms of model functionality and lifecycle management. ML operations (MLOps) offer a systematic approach to tackle these challenges. Existing approaches toward implementing MLOps in a centralized platform often overlook the challenges posed by diverse learning paradigms and network heterogeneity. This article provides a new approach to MLOps targeting the intricacies of future wireless networks. Considering unique aspects of the future radio access network (RAN), we formulate three operational pipelines, namely reinforcement learning operations (RLOps), federated learning operations (FedOps), and generative AI operations (GenOps). These pipelines form the foundation for seamlessly integrating various learning/inference capabilities into networks. We outline the specific challenges and proposed solutions for each operation, facilitating large-scale deployment of AI-Native 6G networks.