[Paper Review] Toward 6G Native-AI Network: Foundation Model based Cloud-Edge-End Collaboration Framework

Xiang Chen, Zhiheng Guo+6 more|arXiv (Cornell University)|Oct 26, 2023

IoT and Edge/Fog Computing4 citations

TL;DR

This paper proposes a foundation model-based 6G native AI framework enabling cloud-edge-end collaboration for task-oriented, intent-aware AI services. By leveraging pre-trained foundation models (PFM) and a unified prompt design, the framework achieves efficient orchestration of wireless resource allocation, demonstrating improved QoS with minimal computational overhead in a multi-access scenario, achieving maximum sum rate with optimized resource consumption.

ABSTRACT

Future wireless communication networks are in a position to move beyond data-centric, device-oriented connectivity and offer intelligent, immersive experiences based on multi-agent collaboration, especially in the context of the thriving development of pre-trained foundation models (PFM) and the evolving vision of 6G native artificial intelligence (AI). Therefore, redefining modes of collaboration between devices and agents, and constructing native intelligence libraries become critically important in 6G. In this paper, we analyze the challenges of achieving 6G native AI from the perspectives of data, AI models, and operational paradigm. Then, we propose a 6G native AI framework based on foundation models, provide an integration method for the expert knowledge, present the customization for two kinds of PFM, and outline a novel operational paradigm for the native AI framework. As a practical use case, we apply this framework for orchestration, achieving the maximum sum rate within a cell-free massive MIMO system, and presenting preliminary evaluation results. Finally, we outline research directions for achieving native AI in 6G.

Motivation & Objective

• To address the limitations of existing AI integration in 6G, particularly the lack of generalizability in task-specific models and scalability issues with large language models at the edge.
• To establish a foundation model-based framework that supports multi-modal, task-oriented AI services across cloud, edge, and end devices.
• To enable efficient, intent-aware AI orchestration in 6G networks through a unified prompt system and collaborative inference.
• To develop a practical implementation of the framework for wireless resource management, demonstrating performance gains in sum rate and computational efficiency.
• To identify open research challenges in evaluation, prompt design, and standardization for future 6G-native AI systems.

Proposed method

• Proposes a 6G native AI framework built on pre-trained foundation models (PFM) to unify data, intelligence, and network collaboration across cloud, edge, and end devices.
• Introduces a customization approach for intent-aware PFM using a unified prompt design to interpret user intents and map them to appropriate AI algorithms.
• Develops a task-oriented AI toolkit that enables dynamic selection and orchestration of AI models based on user intent and QoS requirements.
• Employs a cloud-edge-end collaboration paradigm where the PFM runs centrally with lightweight fine-tuning at the edge and end devices, reducing computational load.
• Utilizes semantic communication principles to enhance data efficiency and reduce transmission overhead in the expert knowledge library.
• Applies the framework to a wireless sum rate maximization task, using the PFM to guide algorithm selection and resource allocation in a multi-access environment.

Experimental results

Research questions

• RQ1 How can a foundation model be effectively customized to support intent-aware, task-oriented AI services in 6G networks?
• RQ2 What collaboration architecture enables efficient, low-latency AI inference across cloud, edge, and end devices in 6G?
• RQ3 How can a unified prompt system be designed to handle diverse, heterogeneous 6G tasks while ensuring robustness and security?
• RQ4 What evaluation metrics are suitable for assessing the performance of intent-aware PFM in complex, multi-objective 6G scenarios?
• RQ5 How can open-source AI standards be harmonized with traditional wireless communication protocols to ensure interoperability in 6G-native AI systems?

Key findings

• The proposed framework achieves maximum sum rate in a multi-access wireless scenario by dynamically orchestrating AI algorithms based on user intent.
• The intent-aware PFM reduces computational resource consumption while maintaining high QoS, as shown in the evaluation results.
• The framework demonstrates effective task selection and resource allocation, with performance improvements validated through simulation-based evaluation.
• The use of a unified prompt system enables accurate intent recognition and efficient orchestration across diverse 6G tasks.
• Preliminary results confirm the feasibility of cloud-edge-end collaboration using PFM for real-time, adaptive network optimization.
• The study identifies critical open challenges in evaluation, prompt robustness, and standardization for future 6G-native AI deployment.