2016 Poster Sessions : Combining Satellite Imagery and Machine Learning to Predict Poverty

Student Name : Neal Jean
Advisor : Stefano Ermon
Research Areas: Artificial Intelligence
Timely and accurate measurements of socioeconomic indicators are fundamental requirements for both sound research and effective policy. However, reliable data on these outcomes remains scant in much of the developing world, slowing efforts to understand the drivers of growth and to implement policies that will improve human livelihoods. We demonstrate an accurate, inexpensive, and scalable method for predicting granular measures of poverty and wealth from high-resolution satellite imagery. Using survey and satellite data from five African countries, we show how a convolutional neural network can be trained to identify informative image features that can then be used to predict localized wealth measures. Our method, which uses only publicly available data, could dramatically improve efforts to track and target poverty in developing countries.

Neal is a second-year PhD student in Electrical Engineering, advised by Stefano Ermon. He is interested in how machine learning can help to solve some of the global challenges facing the world today, such as poverty elimination and climate change.