A conception of mobile edge generation (MEG) is proposed, where generative artificial intelligence (GAI) models are distributed at edge servers (ESs) and user equipment (UE), enabling joint execution of generation tasks. The overall object of MEG is to alleviate the immense network load caused by GAI service and to reduce user queuing times for accessing GAI service. Two MEG protocols are proposed, namely the seed-based MEG protocol and the sketch-based MEG protocol, which enable efficient information exchange and joint generation among ESs and UE. Based on the MEG protocols, various model deployment strategies are proposed to arrange the distribution of the GAI model among UE and ESs. Furthermore, model deployment problems in multi-ES cases are discussed, where several deployment strategies are proposed for parallel and cooperative multi-ES MEG. Finally, a case study, the text-guided- image-to-image generation is provided, where a latent diffusion model is distributed at an ES and a UE. The simulation results demonstrate that both the proposed protocols are able to generate high-quality images at extremely low signal-tonoise ratios, and they also significantly reduce the communication overhead compared to the centralized model. Finally, open research problems for MEG are highlighted.

中文翻译：

---

移动边缘一代：6G 新时代


提出了移动边缘生成（MEG）的概念，其中生成人工智能（GAI）模型分布在边缘服务器（ES）和用户设备（UE）上，从而能够联合执行生成任务。 MEG的总体目标是减轻GAI服务造成的巨大网络负载，并减少用户访问GAI服务的排队时间。提出了两种MEG协议，即基于种子的MEG协议和基于草图的MEG协议，它们能够实现ES和UE之间的有效信息交换和联合生成。基于MEG协议，提出了各种模型部署策略来安排GAI模型在UE和ES之间的分布。此外，还讨论了多ES情况下的模型部署问题，提出了多种并行和协作多ES MEG的部署策略。最后，提供了一个案例研究，即文本引导的图像到图像生成，其中潜在扩散模型分布在 ES 和 UE 处。仿真结果表明，所提出的两种协议都能够以极低的信噪比生成高质量的图像，并且与集中式模型相比，它们还显着降低了通信开销。最后，强调了 MEG 的开放研究问题。