Toward Resource-Efficient Collaboration of Large AI Models in Mobile Edge Networks

TL;DR

A multi-stage diffusion framework that enables elastic distribution of large generative models across heterogeneous edge resources across heterogeneous edge resources is proposed and achieves performance improvement in both efficiency and adaptability for data generation.

Abstract

The collaboration of large artificial intelligence (AI) models in mobile edge networks has emerged as a promising paradigm to meet the growing demand for intelligent services at the network edge. By enabling multiple devices to cooperatively execute submodels or subtasks, collaborative AI enhances inference efficiency and service quality with constrained resources. However, deploying large AI models in such environments remains challenging due to the intrinsic mismatch between model complexity and the limited computation, memory, and communication resources in edge networks. This article provides a comprehensive overview of the system architecture for collaborative AI in mobile edge networks, along with representative application scenarios in transportation and healthcare. We further present recent advances in resource-efficient collaboration techniques, categorized into spatial and temporal approaches. The major spatial approaches include federated tuning, mixture of experts, patch-based diffusion, and hierarchical diffusion. Meanwhile, the important temporal approaches encompass split learning, cascading inference, speculative decoding, and routing inference. Building upon these foundations, we propose a multi-stage diffusion framework that enables elastic distribution of large generative models across heterogeneous edge resources. Experimental results demonstrate that our framework achieves performance improvement in both efficiency and adaptability for data generation.

Figures (4)

• Figure 1: System architecture of collaborative AI in mobile edge networks. The system consists of four layers: the application and demand layer defines service requirements; the collaboration and model layer enables scalable and adaptive model execution; the scheduling and optimization layer coordinates AI tasks and resources; and the device and resource layer provides the underlying infrastructure. Higher layers specify demands, while lower layers offer resource and algorithm support.
• Figure 2: Typical application scenarios of collaborative AI in mobile edge networks. Top: In transportation, roadside sensors collect traffic data, edge servers with lightweight VLMs identify traffic behaviors, and the cloud-based LLM aggregates results for global safety management. Bottom: In healthcare, wearable sensors monitor vital signs, edge devices enable local LLM inference and lightweight finetuning, and the cloud serves as the parameter server for federated finetuning.
• Figure 3: Illustrations of split diffusion and edge-assisted multi-stage diffusion. Top: Split diffusion partitions the model across devices, requiring repeated transmission of feature data at each denoising step. Bottom: Multi-stage diffusion divides the process into contiguous stages, exchanging only latent data between stages to reduce the communication cost.
• Figure 4: Experiments comparison of split diffusion and multi-stage diffusion. (a-b): quality of the generated images under varying resource constraints; (c): proportions of latency incurred by data transmission and computation, respectively; (d) visualization of the generated images by different methods.