Research Projects

Ubiquitous Computing including Wearables and Robotics
Our work in mobile robots and wearable computing focuses on developing resource-efficient algorithms that run directly on smartwatches, fitness trackers, affordable robots, and other battery-powered small devices. We build systems for on-device natural language search, sensor data indexing, continuous health monitoring, and energy-efficient context sensing, all designed to operate within the tight memory and power constraints of wearable platforms.

Digital Health
Our digital health research spans medical image analysis, wearable sensor data mining, computational biology, and global health studies. We’ve developed deep learning models like CovidCTNet for COVID-19 diagnosis from CT images and created diagnostic prediction models for chronic kidney disease. Our work extracts physiological patterns from smartwatch data — analyzing heart rate, physical activity, and behavioral patterns to enable continuous health monitoring and early intervention.

Sustainable Computing
By applying compression techniques and analyzing existing systems in depth, we develop resource-efficient machine learning algorithms that can operate on affordable computers and reduce the carbon footprint of model training. We have applied these techniques to underwater species monitoring, creating shark detection systems for extreme marine environments using domain adversarial neural networks and autonomous camera platforms.

On-device Machine Learning and Edge AI
Our research in Edge AI focuses on resource-efficient algorithms that operate entirely on battery-powered devices without cloud connectivity. We have developed systems for on-device natural language search on wearables, achieving significant improvements in execution speed and energy efficiency. Our work systematically investigates model compression techniques including quantization and pruning across various architectures.

Interpretability and Explainability
By combining guided conversational interfaces, information filtering, and interactive visualizations, we build tools and frameworks that make complex data more accessible and analyzable. One example is a statistical analysis system that uses constrained, step-by-step conversations to guide users toward accurate selection of statistical analysis, outperforming existing state-of-the-art solutions on several benchmark tasks.