Deploying Machine Learning Application on Edge Devices, a Study of Challenges and Solutions

The computing world has seen tremendous explosion of innovation thanks to the development of novel Artificial Intelligence (AI) and Machine Learning (ML) techniques. From AI agents that can win against the most advance human players, to language models that can generate impressive content at superhuman levels. While this is exciting, most of the current research efforts on these fields are focused on developing better algorithms and architectures from the theoretical side, while not much effort is put on understanding the engineering challenges of bringing AI systems to real-life, especially to edge devices. As the community of AI/ML practitioners finds exciting new applications of these techniques we are faced with the challenge of building software systems whose behavior is dictated by data and code. Pioneers on these efforts argue that operationalizing ML systems requires its own engineering discipline (MLOps), and techniques previously applied to build traditional software systems do not cover the complexity of this new type of software system built on data. One exciting real-life application of AI/ML is deploying artificial intelligence on edge devices (phones, smartwatches, cars, security cameras, robots, etc.). This project aims to understand the following aspects of bringing intelligence to edge devices: Identify and motivate key theoretical and practical challenges. Identify and describe proposed solutions (e.g., system architectures). Identify and describe exciting tools (e.g., open-source software, cloud managed-services). Identify and describe best practices and their rationale. Identify and describe systems currently using edge ML. Study economic, ethical, and security aspects of ML systems at the edge.