End-Edge-Cloud Collaborative Computing for Deep Learning: A Comprehensive Survey

The booming development of deep learning applications and services heavily relies on large deep learning models and massive data in the cloud. However, cloud-based deep learning encounters challenges in meeting the application requirements of responsiveness, adaptability, and reliability. Edge-based and end-based deep learning enables rapid, near real-time analysis and response, but edge nodes and end devices usually have limited resources to support large models. This necessitates the integration of end, edge, and cloud computing technologies to combine their different advantages. Despite the existence of numerous studies on edge-cloud collaboration, a comprehensive survey for end-edge-cloud computing-enabled deep learning is needed to review the current status and point out future directions. Therefore, this paper: 1) analyzes the collaborative elements within the end-edge-cloud computing system for deep learning, and proposes collaborative training, inference, and updating methods and mechanisms for deep learning models under the end-edge-cloud collaboration framework. 2) provides a systematic investigation of the key enabling technologies for end-edge-cloud collaborative deep learning, including model compression, model partition, and knowledge transfer. 3) highlights six open issues to stimulate continuous research efforts in the field of end-edge-cloud deep learning.