A Survey of Embedded Machine Learning for Smart and Sustainable Healthcare Applications

Recent advances in machine learning algorithms (ML) and low-power edge devices enable novel wearable applications. Embedded machine learning organically combines these concepts, leading to new application areas. Specifically, machine learning algorithms offer reliable decision-making, classification, and regression performance, while embedded devices allow these algorithms to run at the edge with limited computational power. New embedded devices usually integrate powerful dedicated processors with multiple sensors, such as inertial measurement units. For example, an Nvidia Jetson Nano developer kit is equipped with a quad-core ARM Cortex-A57 processor, 128 CUDA cores with 4 GB memory, and thermal sensors. Hence, embedded machine learning allows performing machine learning directly on devices used in the field, thus leading to numerous novel applications. Promising target applications include health-related applications such as health monitoring, human activity recognition, human pose estimation, and service applications such as energy management in mobile devices.