摘要
端云协同智能计算是大数据、云计算、边缘计算发展的产物,可在保护用户隐私的前提下显著提升数据利用率,实现智能计算实时响应能力与服务鲁棒性的优势互补,而相应技术研发和实践应用具有复杂性.本文剖析了端云协同智能计算的应用价值,凝练了端学习效率优化、端少样本过拟合、端模型定制化、分布差异下虚假关联学习、通信开销与计算效率平衡等方面的技术难题;系统梳理了端云协同智能计算中主流方法研究进展,涉及作为应用基石的高效计算硬件、以端为中心的协同计算、以云为中心的协同计算、端云双向协同计算、可信端云协同智能计算等主要方向;总结了推荐系统、自动驾驶、安防系统、教育模式等端云协同智能计算的垂直领域应用情况.着眼端云协同智能计算的未来发展,需重点研究云资源在端模型个性化中的应用策略、端云协同多目标优化算法、端 ‒ 端与云协同计算的优化策略.
Abstract
Device-cloud collaborative intelligent computing,an emergent result of the development in big data,cloud computing,and edge computing,offers significant improvements in data utilization while protecting user privacy.This approach synergizes the real-time response capabilities of intelligent computing with service robustness.The study explores the application value of this computing paradigm,highlighting technical challenges such as optimizing on-device learning efficiency,mitigating overfitting with limited samples at the device,customizing on-device models,learning false associations under distributional discrepancies,and balancing communication overhead with computational efficiency.We systematically review the progress in mainstream methods within device-cloud collaborative intelligent computing,encompassing efficient computation hardware as the application cornerstone,device-centric collaborative computing,cloud-centric collaborative computing,bidirectional device-cloud collaborative computing,and trustworthy device-cloud collaborative computing.The study also summarizes applications in vertical domains such as recommendation systems,autonomous driving,security systems,and educational models.Looking toward the future of device-cloud collaborative intelligent computing,it underscores the need for focused research on cloud resource application strategies in device model personalization,multi-objective optimization algorithms for device-cloud collaboration,and optimized collaborative strategies between devices and the cloud.
基金项目
科技创新2030—"新一代人工智能"重大项目(2022ZD0119100)
中国工程院咨询项目(2022-PP-07)
国家科技期刊平台
NSTL
万方数据