Study Overview
The mobile phone revolution in low-and-middle income countries has transformed the way humanitarian organizations distribute relief. In the past four years alone, personal data from mobile phones have been used to target economic relief in to individuals experiencing poverty in Togo and Afghanistan. The key insight underlying these approaches is that poorer individuals tend to use their phones differently than richer individuals; hence, when combined with survey data on poverty, techniques from supervised learning can be used to generate a machine learning model that predicts an unknown individual’s poverty status. However, conducting these surveys can be expensive, time-consuming, and oftentimes impossible during a humanitarian crisis. A promising avenue of inquiry would be to collaboratively build a machine learning model using existing data from multiple organizations in unison. However, the use of personal data from economically disadvantaged individuals — even for humanitarian applications — raises privacy concerns.
Study Results
This project will develop tools to foster provably private collaborative machine learning between organizations to support anti-poverty initiatives, and will enable the responsible use of personal data, which in turn can unlock new innovation and facilitate collaborative machine learning between humanitarian organizations.
IBSI Funding Acknowledgement: Lab for Inclusive FinTech (LIFT)