边云协同终身学习面对开放世界:关键技术挑战、探索与案例

Edge-Cloud Collaborative Lifelong Learning for the Open World:Key Technical Challenges,Explorations and Case Studies

由于数据从边缘产生,将部分机器学习任务部署到边缘已成为必然趋势。当机器学习应用逐步被部署到边缘,距离用户更近、面对更复杂环境的开放世界场景将从系统和算法两个角度普遍、持续地影响边缘智能应用的精度和运行。借鉴人类的学习机制,我们已发表一种边云协同终身学习范式以针对上述开放世界中的边缘智能问题。本文总结了边云协同终身学习在理论走向实践过程中遇到的三大技术挑战:可扩展性、任务定义和未知任务,并通过6个工业和园区领域应用案例分享了相关技术探索和经验。

As data is increasingly generated at the edge,deploying machine learning tasks at the edge has become an inevitable trend.When machine learning applications are gradually deployed to the edge,open world scenarios that are closer to users and facing more complex environments will generally and continuously affect the accuracy and operation of edge intelligence applications from both system and algorithm perspectives.Drawing on the learning mechanism of humans,we have published an edge-cloud collaborative lifelong learning paradigm to address the abovementioned edge intelligence problems in the open world.This paper summarizes the three major technical challenges currently encountered in the process of edge-cloud collaborative lifelong learning from theory to practice,i.e.,scalability issue,task definition issue,and unknown task issue.We also share relevant technical exploration and experience through six industrial and campus casestudies.