Distributed Resource Optimization in Network Slicing via Incremental Reinforcement Learning in MEC

With the emergence of fifth-generation (۵G) and sixth-generation (۶G) networks, Network Slicing technology has become a key solution for optimal resource management in shared infrastructures. The automated management of these slices in Edge Computing environments, due to resource distribution and the dynamic nature of requests, presents complex challenges that necessitate advanced Artificial Intelligence techniques. This paper provides a comprehensive analysis of scientific achievements in intelligent and distributed resource management, focusing on Deep Reinforcement Learning (DRL) algorithms. This study investigates ۱۵ selected papers published between ۲۰۱۹ and ۲۰۲۵ from reputable databases and classifies methods based on Markov Decision Process (MDP) modeling, reward functions, and learning architectures. The findings indicate that multi-agent approaches, leveraging the Centralized Training and Distributed Execution (CTDE) framework, lead to a ۱۶% improvement in resource efficiency and a significant ۲۸% reduction in average end-to-end latency. Furthermore, the results confirm that the use of Incremental Learning minimizes the computational cost of model retraining in response to dynamic network changes. Finally, this research shows that despite success in simulation environments, transitioning to real-world testbeds requires improvements in algorithm scalability and stability. The future of this field is moving toward fully autonomous systems through the integration of DRL with distributed structures.

کلیدواژه ها:

Network slicing ، Deep Reinforcement Learning ، Edge computing ، Incremental learning