Enabling Intelligence on Multi-Access Edge Networks with Heterogeneous Resources

日益增长的城市化和物联网（IoT）的日益普及标志着智慧城市时代的到来，这个时代将严重依赖于收集和分析大量数据来改善城市生活。然而，从延迟、成本和安全/隐私的角度来看，通过多个路由器和链路将这些大量分布式数据传输到整个城市的云服务器仍然是一个巨大的挑战。因此，必须在其生成边缘设备（例如传感器、智能手机、监控摄像头、无人机、联网车辆）附近甚至在其生成边缘设备上对尽可能多的这些数据进行分析。由于资源有限，这些异构和无线连接的设备必须作为多访问边缘网络一起工作，以分布式方式执行此类分析。这种在多接入边缘网络上实现智能的创新方向需要革命性的范式，这些范式可以在设备之间共同动态地分配资源和任务，同时考虑到所有计算/网络异构性和能量限制，从而以高精度和/或在所需的时间内完成所有所需的分析。拟议的研究计划将通过为多访问边缘智能（MEI）的新范式奠定基础，在异构边缘设备网络上实现机器学习和数据分析的飞跃。MEI将建立设计要求和实时算法程序，以便在具有异构通信、处理和能量容量的优化/自适应无线边缘设备集群上执行一个或多个（可能相关的）学习/分析任务。它还将涉及物理（即网络/计算）资源、网络形成和学习模型的联合优化，同时适应物理异构性、网络不确定性和移动性考虑，从而实现不同类型应用所需的学习/分析质量指标。此外，该项目将在女王大学建立一个加拿大MEI研究和测试中心，以促进MEI研究，并验证本项目和其他加拿大研究人员和行业开发的MEI创新。培养高素质人才也是该计划的重要和不可或缺的目标。培训内容包括网络设计、边缘计算技术、分布式数据分析、资源优化和MEI的硬件/软件实现。本项目将培养2名博士、2名硕士、2名本科生。鉴于HQP在信息技术、电信、数据分析和边缘情报领域的市场需求非常强劲，该项目训练有素的HQP的未来就业将加速向加拿大工业传播下一代MEI技术。

The increasing urbanization and growing adoption of the Internet of Things (IoT) mark the dawn of the smart cities era, an era that will heavily rely on the collection and analysis of large amounts of data to improve urban life. Yet, transferring these volumes of distributed data to cloud servers across entire cities via multiple routers and links is still a big challenge from the delay, cost, and security/privacy viewpoints. The analysis of as much of this data as possible must thus be performed close to or even on its generating edge devices (e.g., sensors, smartphones, monitoring cameras, drones, connected vehicles). Being limited in resources, these heterogeneous and wirelessly-connected devices must work together as a multi-access edge network to perform such analytics in a distributed manner. This innovative direction of enabling intelligence over multi-access edge networks calls for revolutionary paradigms that can jointly and dynamically allocate resources and tasks among devices, while considering all their computing/networking heterogeneities and energy constraints, so as to fulfill all the desired analytics with high accuracy and/or within the required timelines. The proposed research program will make a leap towards empowering machine learning and data analytics on networks of heterogeneous edge devices by laying the foundations of the novel paradigm of multi-access edge intelligence (MEI). MEI will establish the design requirements and real-time algorithmic procedures that enable the execution of one or multiple (possibly correlated) learning/analytics tasks on optimized/adaptive clusters of wireless edge devices with heterogeneous communications, processing, and energy capacities. It will also involve the joint optimization of physical (i.e., networking/computing) resources, network formations, and learning models, while accommodating physical heterogeneities, network uncertainties, and mobility considerations, so as to achieve the desired learning/analytics quality indicators for different types of applications. In addition, the program will establish a Canadian MEI research and testing hub at Queen's University that promotes MEI research and validates the developed MEI innovations both in this program and by other Canadian researchers and industries. Training highly qualified personnel is also an important and integral goal of this program. This training will include experiences in network design, edge computing techniques, distributed data analytics, resource optimization, and hardware/software implementations of MEI. Two PhD, two MSc, and two undergraduate students will receive training through this research program. Given the very strong market demand on HQP in the information technology, telecommunications, data analytics, and edge intelligence sectors, the future employment of this program's trained HQP will accelerate disseminating next-generation MEI technologies to the Canadian industry.