The Road Toward General Edge Intelligence: Standing on the Shoulders of Foundation Models

Abstract

Foundation models, pre-trained on vast datasets, offer a robust backbone for developing artificial intelligence (AI) applications more efficiently and cost-effectively. Standing on the shoulders of foundation models, that is, by integrating foundation models into edge systems, the system intelligence is poised for significant evolution. This evolution ultimately aims to achieve artificial general intelligence (AGI) at the edge, where edge intelligence (EI) would attain high-level capabilities like reasoning, planning, and learning from experience, potentially reaching or even surpassing human-level proficiency. We refer to this ultimate vision of EI as general edge intelligence (GEI). In this article, we explore the road toward GEI, highlighting the critical role of foundation models in this transformation and offering a glimpse into future application scenarios of GEI. Despite its promising potential, the path is challenged by resource constraints, high energy consumption, inference latency, and privacy concerns. We conducted a survey on recent research directions and efforts aimed at tackling the above challenges, particularly in efficient and safe federated fine-tuning, novel model architectures, and edge system designs for naturally integrating foundation models. By summarizing these advancements and identifying research directions, we aim to provide a broad overview of the progress toward achieving intelligent and responsive edge systems.

Topics & Concepts