Data-centric Edge-AI: A Symbolic Representation Use Case

Abstract

Today’s machine learning pipelines are primarily executed in the cloud, from data storage to data processing, model training, and deployment. However, machine learning is moving to edge devices, creating the demand for AI applications at the edge, known as Edge-AI. Traditional data management practices applied in the cloud are proving to be inefficient for Edge-AI, due to resource and energy constraints of edge devices and real-time requirements of applications. This paper identifies the challenges associated with data processing for Edge-AI. We then discuss methods for efficient data processing at the edge, leading to data-centric Edge-AI. As a use case scenario, we discuss the symbolic representation of time series data and explain how it could help save the cost of data storage and processing in developing Edge-AI applications.