IoT systems consist of a large number of distributed devices such as sensors, actuators, or mobile devices. It has been widely applied for robotics, smart cities, smart agriculture. Numerous end nodes communicate with gateways and cloud servers using various communication protocols including Wi-Fi, BlueTooth, or ethernet.

We study an IoT architecture and application that can be used for controlling the environment, monitoring, and deep learning. We are interested in open-source platforms such as FiWare, and mobile device applications simplifying industrial monitoring.

Deep learning has become a powerful tool for various fields such as autonomous driving, satellite or aerial image processing, medical image processing, or natural language processing. In recent years, many studies have been conducted to implement deep learning network inference on edge devices rather than cloud servers. 

We study the development and evaluation of the deep learning model applied to embedded boards. Since embedded boards generally have lower computational power than mainstream GPU servers, we aim to carefully balance model accuracy, energy efficiency, and computation time.

GPU has been drawn attention as a solution to overcome the performance limitations of conventional multicore-based architecture. However, to achieve the maximum performance of GPU, users need to optimize the parallel program manually. The most critical challenge for optimization is to balance the resource usage of each thread with the number of threads that can be operated simultaneously with the application characteristics.

We study the GPU architecture and implementation for general-purpose computing. Representative applications using GPUs include deep learning, image processing, and edge computing.