Distributed Artificial Intelligence Empowered by End-Edge-Cloud Computing: A Survey

TLDR

The shift from cloud to end‑edge‑cloud computing enables artificial intelligence to transition from a centralized to a distributed paradigm. This survey systematically reviews how distributed artificial intelligence is empowered by end‑edge‑cloud computing, coordinating on‑device, edge, and cloud resources to meet the demands of resource‑intensive AI workloads. It introduces mainstream computing paradigms, the advantages of the EECC model, core distributed AI technologies, a taxonomy of EECC‑driven optimization methods for training and inference, and an analysis of security and privacy threats with corresponding defense strategies. The paper highlights promising applications of DAI‑EECC and identifies key research challenges and open issues for achieving immersive performance acquisition.

Abstract

As the computing paradigm shifts from cloud computing to end-edge-cloud computing, it also supports artificial intelligence evolving from a centralized manner to a distributed one. In this paper, we provide a comprehensive survey on the distributed artificial intelligence (DAI) empowered by end-edge-cloud computing (EECC), where the heterogeneous capabilities of on-device computing, edge computing, and cloud computing are orchestrated to satisfy the diverse requirements raised by resource-intensive and distributed AI computation. Particularly, we first introduce several mainstream computing paradigms and the benefits of the EECC paradigm in supporting distributed AI, as well as the fundamental technologies for distributed AI. We then derive a holistic taxonomy for the state-of-the-art optimization technologies that are empowered by EECC to boost distributed training and inference, respectively. After that, we point out security and privacy threats in DAI-EECC architecture and review the benefits and shortcomings of each enabling defense technology in accordance with the threats. Finally, we present some promising applications enabled by DAI-EECC and highlight several research challenges and open issues toward immersive performance acquisition.