Research
Three connected lines of work: what a night of sleep reveals about health, what a week of wrist movement reveals, and how self-supervised models let those signals transfer between cohorts and devices.
Sleep physiology and digital biomarkers
Extract interpretable markers of sleep health, aging, fragmentation, and neurological disease from polysomnography and related physiological signals.
Automated sleep event detection
Development and validation of deep-learning methods for automated detection of core polysomnographic events, including cortical arousals, leg movements, and sleep-disordered breathing events.
Whole-night sleep biomarkers and phenotyping
Move beyond individual events toward whole-recording and probabilistic descriptions of sleep physiology, including sleep age, fragmentation, and disease-related signatures.
27 publications
Foundation models for physiological signals
Develop and evaluate self-supervised and foundation-model approaches that learn reusable representations across physiological modalities, cohorts, and downstream health tasks.
Foundation models for polysomnography
Large-scale representation learning from multimodal sleep recordings for transferable health and disease prediction — spanning the model itself, the pretraining methodology behind it, and work interpreting what it learns.
Foundation models for wrist accelerometry
Self-supervised and physiology-aware representation learning from wrist accelerometry, spanning human activity recognition, sleep staging, apnea evaluation, and future health prediction.
9 publications
Wearables for scalable health assessment
Use wrist accelerometry and other low-cost sensors to derive scalable physiological markers for sleep, activity, neurodegenerative disease screening, and future health risk.
Wearable screening for isolated REM sleep behavior disorder
A progression from single-cohort proof of concept to multicenter, device-agnostic validation, a practical two-stage screening protocol, and richer physiological signatures beyond simple actigraphy.
Gait and motor function from wrist accelerometry
Detecting walking bouts from free-living wrist accelerometry, then using that same signal to characterize gait alterations associated with neurodegenerative disease.
Accelerometry and future health risk
Foundation-model architectures for wrist accelerometry applied to population-scale prediction of future disease risk, and to differentiating dementia etiologies in a memory-clinic population.
15 publications