Large-scale School Mapping using Weakly Supervised Deep Learning for Universal School Connectivity

AI can help identify schools in underserved regions: The researchers achieved over 96% accuracy (AUPRC values above 0.96) in locating schools across 10 African countries using satellite imagery and machine learning, demonstrating that AI can help governments and organizations find schools that may be missing from official records—critical data for planning connectivity infrastructure.

Weak supervision makes AI solutions more practical and affordable: Rather than requiring expensive, detailed manual annotations of every school location, the team used "weakly supervised" techniques that work with simpler, lower-cost labeling. This approach makes the solution more scalable and accessible for low- and middle-income countries with limited resources.

Explainable AI helps build trust with stakeholders: The researchers used explainable AI techniques to show why their model identified certain locations as schools, making the results more transparent and trustworthy for government partners who need to validate and act on these findings.

Human oversight remains essential: The team built an interactive web mapping tool specifically designed for "human-in-the-loop" validation, recognizing that AI predictions should be reviewed and verified by human experts before being used for real-world infrastructure planning decisions.

Combining different AI techniques improves results: The best-performing models merged vision transformers with convolutional neural networks, illustrating that responsible AI implementation often involves thoughtfully combining multiple approaches rather than relying on a single method.