I. Introduction
With the rise of Automatic Machine Learning (AutoML) techniques, significant progress has been made to automate the design of deep learning (DL) architectures. Among these techniques, Hardware-aware Neural Architecture Search (HWNAS) [1], [2] has recently gained a lot of attention from the DL community. HW-NAS takes NAS a step further by automating the design of DL architectures on resource-constrained devices.