AI-SDM Seminar - Tom Manzini
— 1:30pm
Location:
In Person and Virtual - ET
-
Newell Simon 4305 and Zoom
Speaker:
TOM MANZINI
,
Ph.D. StudentDepartment of Computer Science & Engineering Texas A&M University
https://www.nzini.com/
When a disaster strikes, disaster response operations deploy multiple sensors to capture imagery of impacted areas so decisions can be made with the best information possible. Satellites, small uncrewed aerial systems (sUAS), and crewed aircraft are deployed to capture imagery of the affected area. However, once that imagery is captured, it arrives in waves, creating data avalanches that can paralyze decision-making and thus the disaster response. Computer vision models are needed to manage this immense volume of data, but in practice, it is impossible to know which sources will be available when decisions need to be made: clouds can obscure the scene from satellites, weather may prevent crewed aircraft from flying, and sUAS teams may be physically unable to access the disaster scenes. To manage this situation, computer vision systems that can accurately and gracefully handle this diverse real-world imagery are critical, but such systems do not yet exist.
This talk will explore the research and fieldwork behind the development and deployment of multi-source, multi-scale damage assessment systems. It will introduce CRASAR‑U‑DROIDs, the largest dataset of its kind, spanning 10 disasters, 3 imagery sources, 70.6 gigapixels, and 122,502 labels, and discuss the technical and operational challenges of bringing these systems to real‑world disaster environments. The presentation will continue by covering the first known operational deployment of an sUAS-based automated damage assessment system during Hurricanes Debby and Helene, and the talk will conclude by discussing current academic efforts to enhance model cross-scale capabilities to strengthen decision-making during disasters.
—
Thomas Manzini is a PhD student at Texas A&M where his work focuses on computer vision and machine learning, specifically in increasing the robustness and applicability of machine learning to support operational decision-making like that found in wide-area disaster response. Thomas has a B.S. from Rensselaer Polytechnic Institute and an M.S. from Carnegie Mellon University. He has helped organize the AI for Humanitarian Assistance and Disaster Response workshop at NeurIPS and ICCV. Before returning to academia for his PhD, Thomas worked as a Machine Learning Scientist at Microsoft, where he split his time between machine learning products and collaborations with groups like the CDC and the WHO in the development of machine learning and data management systems to respond to disasters like the COVID-19 pandemic. Thomas holds Commercial Pilot Licenses for Manned (FAA Part 61) and Unmanned (FAA Part 107) aircraft and has more than a decade of operational experience as an Advanced Emergency Medical Technician and Firefighter.
REGISTER → to attend in-person or on Zoom
For More Information:
pwerns@andrew.cmu.edu