The TR&D1 team (Dr. Jim Rehg, Dr. Santosh Kumar and Dr. Benjamin Marlin) is dedicated to identifying optimal moments when health risks are heightened yet manageable, leveraging innovative technologies to delve deeper into the nexus of health and behavior. Here’s a glimpse into our recent endeavors and opportunities for collaboration:

Advancing Health Behavior Understanding: Our research thrust investigated how to enhance decision-making in administering precise mHealth interventions. By grappling with the complexities of longitudinal mHealth data, the Center has been pushing the boundaries of understanding health behavior dynamics.

Imputing Missing Data with Precision: One of our focal points has been the development of robust imputation methods for missing step count data from Fitbit devices. With over 10,000 participants and millions of hourly step count observations across two trials, our custom self-attention-based model stands as a robust foundation model for step count imputation and forecasting.

Pioneering Deep Learning for Biophysical Signals: Beyond step counts, we’ve delved into the realm of pulsatile biophysical signals like ECG and PPG. Leveraging uncertainty modeling and a transformative Bottleneck Dilated Convolutional (BDC) architecture from the PulseImpute dataset, we’ve charted new territories in imputation and forecasting, laying the groundwork for future breakthroughs.

Cutting-edge Self-Supervised Learning Approach:  By leveraging reconstruction error in imputation within a contrastive learning framework, especially for pulsatile waveforms, we’ve broken down barriers in understanding sparse labels in mHealth signal data. Our proven training and embedding functions deliver remarkably effective feature representations for downstream tasks, offering unparalleled opportunities for impactful collaborations.

To learn more about the mDOT Center’s TR&D1 and the various research outcomes, visit TR&D1: Discovery.