CURRENT RESEARCH

Machine Learning based Algorithms and Software for Transportation and Smart Cities

My group develops machine learning algorithms and software for analyzing traffic and improving intersection safety and operations. We combine video, LiDAR, and ground-sensor data to detect and track vehicles and pedestrians, reconstruct trajectories, and identify potential collisions and near-misses. Our traffic-modeling research includes recurrent attention networks, graph attention networks, digital twins, and generative models for intersection and corridor dynamics. These methods support traffic-flow prediction, simulation calibration, measures of effectiveness, and signal-policy optimization.

Data Science for Health Care Applications

My health-care research combines wearable sensing, cloud software, and machine learning to study activity, mobility, and health outcomes. A central project is ROAMM, the Real-time Online Assessment and Mobility Monitor, developed with collaborators in aging and health research. ROAMM uses a smartwatch to support continuous measurements and bidirectional interaction with participants. Programmable measurements and participant-reported outcomes provide data surrounding falls, injuries, hospitalizations, and other intervening health events. Studies capture activity and mobility patterns together with symptoms such as pain and anxiety.

Hybrid Multicore and Exascale Computing for Scientific Applications

My group develops algorithms and software for multicore CPUs, GPUs, and heterogeneous computing systems. We focus on applications with sparse data and changing computational structure, including compressible multiphase turbulence, synthetic aperture radar reconstruction, and sparse matrix factorization. The research addresses task-to-core mapping, dynamic load balancing, communication and synchronization, memory organization, and energy-aware scheduling. Our methods consider both computational performance and energy requirements, including dynamic voltage and frequency choices and reconfigurable caches.

Machine Learning for Data Reduction of Scientific Applications

Scientific simulations generate data at scales that challenge storage, communication, and analysis. My group develops machine learning and hybrid compression methods that reduce scientific data while controlling reconstruction error and preserving downstream quantities of interest. Recent work extends these methods to foundation and generative models for spatiotemporal scientific data, including CAESAR and generative latent diffusion. The research studies how learned representations and correction methods can provide useful reduction while meeting specified scientific error requirements.