Towards Online Learning and Concept Drift for Offloading Complex Event Processing in the Edge

Abstract

Edge computing has enabled the usage of Complex Event Processing (CEP) closer to data sources, delivering on time response to critical applications. One of the challenges in this context is how to support this processing and keep an optimal resource usage (e.g., Memory, CPU). State-of-art solutions have suggested computational offloading techniques to distribute processing across the nodes and reach such optimization. Most of them take the offloading decision through predefined policies or adaptive solutions with the usage of machine learning algorithms. However, these techniques are not able to incrementally learn without any historical data or to adapt to changes on statistical data properties. This research aims to use online learning and concept drift detection on offloading decision to optimize resource usage and keep the learning model up-to-date. The feasibility of our approach was noticed through preliminary evaluations.