Sleep research is a translational science field, and can greatly benefit from improved computational technologies. Automated, robust, interpretable, and high-fidelity models are vital to its success. The pervasive adoption of wearable devices provides a unique opportunity. Wearable devices monitor a user for an extended period of time and can generate a large amount of data. The current sleep analysis processes are manual and unable to scale, creating a bottleneck for sleep research. Automation allows for immediate analysis on large-scale clinical trials, and provides a platform for affordable widespread population screening of sleep disorders. Particularly for population screening, it is critical that the methods and knowledge extracted are generalizable and thus robust to noise and variance amongst sub-populations. For example, teenagers follow very different sleep patterns compared to the elderly. Moreover, sleep experts use software tools such as ActiLife to manually annotate datasets. This leaves the data interpretation prone to human error.
This talk presents a translational science approach to human health through sleep analysis, creating novel state-of-the-art computer science algorithms to empower clinicians and patients alike. We introduce the basics of sleep science, highlighting the computing challenges in the field, and proposing computational solutions to improve the sleep science process. By providing tools for sleep and activity behavior analysis, clinical decision- makers can deliver improved and informed healthcare. Moreover, this research also empowers patients from a quantified-self perspective, by conveying real-time recommendations to optimize productivity and improve quality of life.