Edge Large AI Model: Principles and Applications
Date:
Large artificial intelligence models (LAMs) have demonstrated human-like capabilities in addressing a broad spectrum of real-world problems, showcasing transformative potential across diverse domains and modalities. In contrast to traditional edge AI systems—typically designed for single-task execution using compact models—edge LAMs require the decomposition and distributed deployment of large-scale models, while enabling support for highly generalized and multifaceted tasks. However, the deployment of edge LAMs faces significant challenges due to the limited communication, computation, and storage resources inherent in wireless edge environments. The enormous number of trainable parameters and the associated communication overhead create substantial barriers to practical implementation. In this talk, we explore the key opportunities and fundamental challenges in realizing edge LAMs, with a focus on model decomposition strategies and intelligent resource management. We further discuss enabling technologies that facilitate large-scale deployment and outline promising research directions in the context of future 6G networks.