PhD Thesis Defense: Priyanka Mary Mammen, Enabling Scalable Sleep Monitoring with Mobile Sensing and Machine
Content
Speaker
Abstract
Sleep is a critical component for the overall health of a human being. Despite its importance, a majority of the population is sleep deprived, leading to several physical and mental health issues. Traditional methods of sleep monitoring are expensive and not scalable, limiting access to important health information. The advent of the Internet of Things and mobile sensing devices provides an opportunity for more accessible and scalable sleep monitoring. Community-scale sensing has the potential to enable aggregate public health monitoring and informed decision-making for individuals. However, scaling mobile health sensing technologies to the community level presents several challenges, including data availability, model generalizability, and robustness.
This thesis focuses on addressing the challenges of sleep monitoring at the community level by developing scalable, personalizable, and robust sleep detection techniques. The first technique, called WiSleep, a system that utilizes an unsupervised machine learning approach that detect sleep durations from WiFi activity of mobile devices. WiSleep leverages the strong correlation between a phone’s network activity and sleep periods and uses an ensemble of Bayesian models designed to handle irregular sleep patterns. The second technique, called SleepLess, is a semi-supervised machine learning approach that enables personalized sleep estimations for users without labeled data. The third technique, called SleepWell, is a post-hoc error correction method for automated sleep stage classification, combining novel uncertainty quantification metrics with rule-based techniques. Overall, this research aims to provide innovative solutions to transition from reactive to proactive health sensing, facilitating early detection and intervention in sleep-related health conditions.
Advisor
Prashant Shenoy