Weak Supervision Facilitates Mapping Wealth and Population Using Historical Aerial Photographs

In this month's forum, ​Dr. Joel Ferguson (Sustainable Land Systems, Nelson Institute) presents an applied ML pipeline that links historical aerial imagery to long-run economic outcomes. 

Please fill out the registration form by EOD Monday (Glass Nickel Pizza provided in-person). Hope you can join us!

When: Tuesday, March 10, 12-1pm CT
Where: Orchard View, Discovery (and Zoom - passcode 111195)

Abstract: To understand how best to promote economic growth, alleviate poverty, and reduce inequality, we need measures of wealth that are both disaggregated over space and available over long time periods. Reliable sub-national data on wealth only go back to 1990 for many developing countries, meaning that many important policy questions cannot be answered using existing data. To begin to fill this gap, we apply frontier machine learning methods to a newly digitized archive of 1.3 million historical aerial photographs to produce a novel 5 km resolution dataset of population density and wealth per capita for 18 African countries covering the period from 1940-1990. Our machine learning pipeline combines transfer learning, incorporating modern satellite imagery and labels, with weak supervision using subjective pairwise rankings of historical images generated by human annotators. Our models prove skillful in both predicting modern outcomes and ranking historical outcomes, and we validate historical population predictions using newly digitized and georeferenced census data. We use our newly developed predicted wealth and population measures to decompose long-run per capita growth into urban growth, rural growth, and urbanization components.

Finding Orchard View: The Orchard View room is located on the 3rd floor of Discovery Building—room 3280. To get to the third floor, take the elevator located next to Aldo’s Cafe kitchen (see photo). 

Join ML+X / Share Your Work! This talk is part of a monthly forum hosted by the ML+X community at UW-Madison. Join our Google group to be notified of future events. Better yet, sign up to discuss your ​ML/AI work at the monthly forum! The majority of our attendees are applied practitioners from diverse fields, not AI/ML purists looking to critique. Contact endemann@wisc.edu if you have any questions.