Collaborative Fine-Tuning of Mobile AIGC Models with Wireless Channel Conditions

Fine-tuning mobile artificial intelligence generated content (MAIGC) models is a crucial procedure for delivering personalized services to mobile users. However, how to reasonably share the computational complexity, and protect user privacy through the collaboration of edge devices and mobile users under wireless channel conditions, is a problem that must be solved in MAIGC. In this article, we explore a collaborative fine-tuning MAIGC pattern based on the forward diffusion process, taking into account the wireless channel characteristics and content features. The input image features and sampled noise features of content are exploited for the forward diffusion process, and superimposed by over-the-air computation (AirComp). Meanwhile, the wireless channel noise is used as the sampled noise for noise recycling, which improves the fine-tuning efficiency, and reduces the power consumption at the same time. Finally, we highlight open research challenges and provide a concise summary.