优化移动边缘AI生成一切(AIGX)服务:提示工程的原理、框架与案例研究
Optimizing Mobile-Edge AI-Generated Everything (AIGX) Services by Prompt Engineering: Fundamental, Framework, and Case Study
AI总结:
本文提出移动边缘AIGX概念,构建统一框架利用边缘设备提供PFM赋能的AIGX服务,并通过提示工程优化服务,案例研究证明有效提示优化可提升用户体验、生成质量与网络性能。
摘要:
作为下一代内容创作范式,AI生成内容(AIGC),即基于用户提示由生成式AI(GAI)自动生成内容,近来已获得广泛关注并取得成功。随着GAI能力的不断增强,尤其是包含数十亿参数的预训练基础模型(PFMs)的出现以及提示工程方法(即为给定任务寻找最佳提示)的发展,AIGC的应用范围正迅速扩大,覆盖了面向人类、系统和网络的各种信息形式,如网络设计、信道编码和优化解决方案。在本文中,我们提出了移动边缘AI生成一切(AIGX)的概念。具体而言,我们首先回顾了AIGX的构建模块、从AIGC到AIGX的演变以及实际的AIGX应用。然后,我们提出了一个统一的移动边缘AIGX框架,该框架利用边缘设备提供基于PFM的AIGX服务,并通过提示工程优化这些服务。更重要的是,我们证明了次优的提示会导致较差的生成质量,进而对用户满意度、边缘网络性能和资源利用率产生不利影响。据此,我们进行了一项案例研究,展示了如何使用ChatGPT训练一个有效的提示优化器,并调查了提示工程在用户体验、生成质量和网络性能方面能带来多大的改进。
英文摘要:
As the next-generation paradigm for content creation, AI-Generated Content (AIGC), i.e., generating content automatically by Generative AI (GAI) based on user prompts, has gained great attention and success recently. With the ever-increasing power of GAI, especially the emergence of Pretrained Foundation Models (PFMs) that contain billions of parameters and prompt engineering methods (i.e., finding the best prompts for the given task), the application range of AIGC is rapidly expanding, covering various forms of information for human, systems, and networks, such as network designs, channel coding, and optimization solutions. In this article, we present the concept of mobile-edge AI-Generated Everything (AIGX). Specifically, we first review the building blocks of AIGX, the evolution from AIGC to AIGX, as well as practical AIGX applications. Then, we present a unified mobile-edge AIGX framework, which employs edge devices to provide PFM-empowered AIGX services and optimizes such services via prompt engineering. More importantly, we demonstrate that suboptimal prompts lead to poor generation quality, which adversely affects user satisfaction, edge network performance, and resource utilization. Accordingly, we conduct a case study, showcasing how to train an effective prompt optimizer using ChatGPT and investigating how much improvement is possible with prompt engineering in terms of user experience, quality of generation, and network performance.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。