PhD Thesis Defense: Forsad Al Hossain, Towards Privacy-Sensitive Edge-Based Crowd and Syndromic Signal Monitoring Contactless Systems
Content
Speaker
Abstract
Recent advances in deep learning and AI have opened new pathways for crowd and syndromic signal monitoring. However, building practical applications in the real world while maintaining privacy, security, and accuracy remains a challenge. Additionally, building syndromic signal monitoring platforms is an unexplored area of research. To address these issues, we have developed multiple novel, first-of-their-kind approaches using edge-AI for crowd and syndromic signal monitoring.
First, we present our work, FluSense, where we built the world’s first syndromic edge-AI system for flu-outbreak monitoring. We demonstrate how to build a practical real-world system that can collect sensor data from a hospital waiting room setting and convert the collected data to predict the total daily flu patient visits in that hospital waiting room. We will also discuss our follow-up work on FluSense, where we showed how a contactless sensing platform can be utilized for COVID outbreak monitoring and how it can outperform traditional thermal scanner-based screening methods for public health monitoring.
Additionally, We will also discuss our work where we show how occupancy can be estimated in a public area using non-speech environmental audio sounds and we will also discuss how systems built on top of FluSense can be utilized for air purification purposes.
Finally, we will present our work on privacy-preserving deep learning models that leverage novel deep learning training methodologies. Our research introduces innovative training methodologies that simultaneously enhance privacy protection and model interpretability. We demonstrate that these approaches are highly versatile, scaling effectively beyond audio applications to diverse domains including computer vision and physiological symptom monitoring.
Advisor
Tauhidur Rahman