On average each family today owns 10 connected smart devices, which will rise to 50 by 2020. Meanwhile, breakthroughs in AI technologies hold the promise of transforming several industries. Together, they can bring unprecedented scale of physiological, behavioral, and cognitive digital markers for Alzheimer’s disease. However, the prominence of smart devices inevitably leads to higher risks of privacy violation and data leakage. Moreover, the increased number of digital markers and the “black box” nature of many AI algorithms make it extremely difficult to interpret the correlations of different digital markers and their links with disease pathophysiology.

Comprising experts in sensor systems, Internet of Things, AI, human-centered computing, nursing, psychiatry, Alzheimer’s disease and dementia, our multi-disciplinary team aims to develop privacy-preserving deep learning technologies for off-the-shelf smart devices for Alzheimer’s disease. The proposed deep learning technologies include 1) a suite of algorithms to classify digital biomarkers for Activities of Daily Living (ADL), Behavioral and Psychological Symptoms of Dementia (BPSD), social interactions, motor function, and cognition, based on a combination of motion sensors in wearables, acoustic sensors in mobile/home devices such as tablets, and depth sensors (standalone or embedded on cameras of latest smartphones)  that only extract 3D distances and hence preserves user privacy; 2) a real-time federated learning system which enables local algorithms collaboratively improve learned models while keeping all the training data on device; and 3) interpretable deep learning algorithms to quantify the correlation of multi-modal digital biomarkers, and assist early detection, diagnosis and intervention. Our technologies will be deployed and validated through a cohort of totally 200 patients.