RITA: Reliable and Efficient Task Management in Edge Computing for AIoT Systems

点击翻译按钮获取中文摘要

Artificial Intelligence of Things (AIoT) has been attracting significant research attention worldwide owing to its great potential in boosting the promising AI-based innovations to support emerging smart applications, such as intelligent manufacturing, autonomous driving, and smart city. To handle massive volumes of data and process computation-intensive AI algorithms, edge computing has emerged as a key building block to empower AIoT. The lifecycle of computational task management in AIoT consists of task offloading decision, task transmission, and task execution. To tackle these challenges, this project aims to create a suite of novel theories and approaches of task management towards achieving reliable and efficient edge computing for AIoT systems. To this end, I will firstly develop an original parallelism- and dependency- aware offloading decision scheme to find the fine-grained tasks with maximised parallelism and minimised dependency, such that more effective task workloads can be offloaded with less energy consumption. I will then propose a novel task transfer protocol that adopts a lightweight cross-access link prediction scheme based on the noise patterns. Finally, an opportunistic task allocation scheme will be designed, which allocates each task to multiple opportunistic and redundant edge servers for achieving the minimum delay for task execution. A holistic AIoT testbed and open-access software will be implemented for performance evaluation and demonstration of the proposed task management approaches. The research outcomes will contribute to Europe's competitiveness and growth in the promising AIoT area. Well-planned training activities will equip me with new skills and competences to enhance my creative and innovative potential and promote my career prospects. Open science practices as well as planned communication, dissemination, and exploitation activities will reach out to society at large and make the research results visible to citizens.