Population Dynamics and Building Occupancy

Settlement and Building Detection in Satellite Imagery

• Managed a project team of 9 scientists to scale up CNN-driven settlement detection and building detection workflows to full country scale. -Developed a data workflow from local clusters and DGX boxes to deployment of deep learners on Titan.
• Developed and optimized GPU-based image tiling technique to integrate in deep learning workflow (pycuda).

Building Occupancy from Simulated Traffic

• Solicited traffic data from Illinois Department of Transportation (IDOT) for traffic models.
• Implemented coupling between population and transportation mode choice models and traffic simulations (TRANSIMS/SUMO) to generate realistic transit scenarios for an area of 34sqm in the Chicago metropolitan area.
• Implemented coupling between traffic simulation and building energy models (EnergyPlus). -Designed and set up realistic, data-driven traffic scenarios for TRANSIMS.
• Ran parallel TRANSIMS simulation models on Titan supercomputer.

Building Occupancy from Traffic Sensor Measurements

• Processed and managed traffic counts from over 30 IoT traffic monitoring devices.
• Mapped the device topology to geographic locations and produced a Voronoi tesselation to assign the percentage of arriving and departing traffic to 600 buildings.
• Featured in STEM Magazine and ACM TechNews.

Selected Publications:

• Performance Analysis for Country-Scale Settlement Detection
• Hyperparameter Optimization for CNNs using Evolutionary Algorithms
• Building Occupancy from Simulated Traffic
• Analysis of Weather Impact on Speed Patterns
• Building Occupancy from Traffic Sensors