As deep learning for resource-constrained systems become more popular, we see an increased number of intelligent embedded systems such as IoT devices, robots, autonomous vehicles, and the plethora of portable, wearable, and mobile devices that are feature-packed with a wide variety of machine learning tasks. However, the performance of DNNs (deep neural networks) running on an embedded system is significantly limited by the platform’s CPU, memory, and battery size; and their scope is usually limited to simplistic inference tasks only. This tutorial introduces on-device deep learning algorithms and supporting systems designs, which enable embedded systems to efficiently perform deep intelligent tasks, deep neural networks in particular, beyond their limited computing resources. We name such on-device deep intelligence on embedded systems as Embedded Deep Intelligence. Specifically, we propose resource-aware learning strategies devised to overcome the fundamental constraints of embedded systems. Once deployed in the field with the proposed resource-aware learning strategies, embedded systems are not only able to perform deep inference tasks on sensor data but also update and re-train their learning models at run-time without requiring any help from any external system. Such an on-device learning capability of Embedded Deep Intelligence makes an embedded intelligent system real-time, privacy-aware, secure, autonomous, untethered, responsive, and adaptive without concern for its limited resources.

In most of the research studies of communication systems, stationarity of traffic is assumed, and the associated processes are approximated to some known "standard" distributions. While such assumptions are indeed necessary for developing tractable analytical frameworks for performance evaluation, at times such assumptions are far from reality. Moreover, in modern-day IoT communications, for resource efficiency more precise optimizations are necessary, where the assumed stationarity and traffic distributions prove to be quite strong and do not necessarily result in predicting accurate performance trends. Therefore, more and more researchers are resorting to data-driven dynamic system characterization and optimization strategies. In this context, in this discourse we will present data-driven approach to performance optimization of communication systems. Through a few examples, which include cognitive radio spectrum access, smart power grid monitoring, smart sensing IoT systems, we will demonstrate how data-driven, context-aware light-weight machine learning based, and edge computing aided approaches are utilized in more accurate system performance characterization and optimization leading to communication and storage resource efficiency and energy sustainability of the IoT nodes. We will share our experiences from field experiments, proof-of-concept implementations, and deployments, and will highlight the possibilities of learning-aided networking optimizations for various multidisciplinary-disciplinary applications.