AI Design in mHealth Systems

The utility of AI in mobile health (mHealth) systems is rapidly growing due to the potential for objective and remote health monitoring and prediction of life-critical events. However, the adoption of emerging machine learning models such as neural networks for use with these systems is challenging. On one hand, embedded devices are limited in compute power, memory storage, and energy source, thus preventing data-intensive and computationally complex models from real-time execution on such edge devices. On the other hand, the distribution of the sensed data changes over time as the system is being deployed in new environments, adopted by new users, or begins to learn new tasks. The distribution shift results in drastic performance decline of the AI models. This warrants the development of new approaches that offer robust machine learning under constant distribution shift. The talk will present our recent research on designing AI technologies for efficient and robust machine learning and their applications in mHealth. We will discuss how AI solutions can be designed for diabetes prevention and management, human-in-the-loop learning, continual learning, and deployment on edge devices.

Bio
https://ghasemzadeh.com/