Advanced Topics in Edge Sensing and Learning

The course covers theoretical, system, and application aspects pertaining to the use of mobile, wearable, and the Internet of Things devices (herefrom referred to as “edge devices”) for sensing and learning about the environment. The course starts with the overview of edge sensing platforms, covering topics such as the constraints and applications of these platforms, and the functioning of these platforms, thus touching upon the sampling theory, including the recent advances in sub-Nyquist sampling - compressive sensing. The course then focuses on deep learning (DL) on edge devices, more specifically, on advanced topics, such as dynamic model compression techniques and running contemporary large language models (LLMs) on mobile devices. We then discuss distributed machine learning training on edge devices through split and federated learning. The course then presents in-depth investigation of applications of DL on edge devices, e.g. for healthcare, authentication and security. Finally, we present a critical analysis of sustainability in edge computing. A key component of the course is a practical project that students will independently work on. The project harnesses modern tools for mobile sensing (e.g. Android) and on-device deep learning (e.g. TensorFlow Lite) and requires students to develop either a full-fledged edge deep learning application or generate new insights from pre-collected sensor data. Student participation is facilitated further by mandatory research paper presentations that will be delivered by each student in the class.

Keywords: Mobile computing, Edge computing, Mobile deep learning, Federated learning, Split learning, mHealth, Mobile security, Computing and sustainability, Smartphones, Wearables, Internet of Things (IoT), Android