Knowledge Transfer and Reuse: A Case Study of Ai-Enabled Resource Management in RAN Slicing

Abstract

Reconfigurable intelligent surfaces (RISs) have received considerable attention as a key enabler for envisioned 6G networks, for the purpose of improving the network capacity, coverage, efficiency, and security with low energy consumption and low hardware cost. However, integrating RISs into the existing infrastructure greatly increases the network management complexity, especially for controlling a significant number of RIS elements. To unleash the full potential of RISs, efficient optimization approaches are of great importance. This work provides a comprehensive survey on optimization techniques for RISaided wireless communications, including model-based, heuristic, and machine learning (ML) algorithms. In particular, we first summarize the problem formulations in the literature with diverse objectives and constraints, e.g., sum-rate maximization, power minimization, and imperfect channel state information constraints. Then, we introduce model-based algorithms that have been used in the literature, such as alternating optimization, the majorization-minimization method, and successive convex approximation. Next, heuristic optimization is discussed, which applies heuristic rules for obtaining low-complexity solutions. Moreover, we present state-of-the-art ML algorithms and applications towards RISs, i.e., supervised and unsupervised learning, reinforcement learning, federated learning, graph learning, transfer learning, and hierarchical learning-based approaches. Modelbased, heuristic, and ML approaches are compared in terms of stability, robustness, optimality and so on, providing a systematic understanding of these techniques. Finally, we highlight RISaided applications towards 6G networks and identify future challenges.

Distributed Machine Learning and Native AI Enablers for End-to-End Resources Management in 6G

“…Here, our aim is to concentrate on a subset of methodologies that are more pertinent to the 6G vision, in alignment with the challenges discussed in the previous section. Specifically, using as a starting point existing surveys, we examine the targeted subset of methodologies and assess their suitability for 6G in terms of the following benchmarking criteria: (a) Whether they are scalable to larger topologies [1] (scalability); (b) Whether they can become a part of a larger end-to-end framework for RM [4,14,16] (composability/modularity); (c) Whether their accumulated knowledge can be extracted and reused [11] (transferability).…”

“…To complement the combined viewpoint targeted by the article, we then turn to network-native AI enablers that are part of the 6G vision. Specifically, we consider three complementary enablers: the AI plane that establishes AI workflows to manage data collection and status monitoring and perform AI model training, planning, and AI model deployment to the network [1,10]; knowledge reuse facilities that allow for model libraries, model selection, and transfer learning [11]; and digital twin networks (DTNs), that will enable network performance prediction and model training [12,13]. We describe how these enablers support the function of AI models in general, but also, importantly, discuss the potential of the enablers and the implications involved when they provide support for each of the six aforementioned distributed RL frameworks.…”

“…Nevertheless, there are a few earlier works addressing some aspects of end-to-end RM in conjunction with some notion of AI enablers, often coming under the label of pervasive AI (PAI), and we now comment upon such related work. Publication [11] discusses end-to-end RM in the narrower context of slicing, and transfer learning is examined as an enabler for the application of ML in next-generation networks. A taxonomy of AI methodologies is provided, together with a discussion of the applicability and limitations associated with these methodologies.…”

“…There are various ways to store knowledge, including as trained models, collected experience, or exploration strategies in the context of RL, organized by associated task. Mapping functions from source tasks to target tasks can even create learner models that are able to execute tasks jointly [11]. Knowledge reuse will be embedded into AI workflows, and its data collection intervals and storage will be controlled by the AI plane.…”

Transfer Learning-Based Accelerated Deep Reinforcement Learning for 5G RAN Slicing

“…These approaches can help pave the way for adopting trustworthy DRL to optimize dynamic RRM functionalities in O-RAN. Transfer learning (TL) is among the main methods used to address the DRL-related practical challenges mentioned earlier [14], [15]. TL can be used to guide a newly deployed DRLbased xApp while learning the optimal policy in network conditions it has not experienced before.…”