We aim at developing an agile Edge Intelligence framework based on methodology of co-designing communication, computing and machine learning models to maximize service reliability, i.e., machine learning inference accuracy within a delay constraint, for delay-sensitive IoT applications.

The emerging Internet of Things (IoT) applications, e.g., AR/VR, digital twins, are delay sensitive. The current solutions cannot meet the latency and reliability requirements. Mobile Edge Computing (MEC) has potential to address the challenges. However, the static machine learning (ML) models do not allow MEC to adapt ML process based on time budget; thereby, it cannot provide accurate inference with a guaranteed latency.

In the AgilE-IoT project, we aim to develop an agile MEC framework integrating ML models of adjustable inference time by co-design of communication, computing and ML model. It enables collaborative inference of IoT devices and edge servers to maximize the inference accuracy within a delay constraint. The results will serve as a foundation for developing diverse delay sensitive IoT applications and have a profound impact on the digitalized society.