A new artificial intelligence technology can identify impoverished urban areas through satellite imagery, potentially transforming urban policy development in developing nations.
KAIST announced Monday that a research team led by Professor Cha Mi-young of the School of Computing and Professor Kim Ji-hee of the School of Business and Technology Management developed the satellite-based universal slum detection AI in collaboration with Professor Yang Jae-seok of the Geography Department at Chonnam National University.
Urban poverty zones have posed challenges for data collection due to significant variations in building shapes and density, resulting in low accuracy. The problem is particularly acute in developing countries, where data marking the locations of impoverished areas is scarce, making AI training difficult.
To address these challenges, the research team implemented a "Mixture-of-Experts" architecture, where multiple AI models learn characteristics of different regions and automatically select the most suitable model when a new city is input. The team also applied "Test-Time Adaptation" technology, enabling the AI to improve its own performance in new cities by comparing prediction results from multiple models and trusting only areas where results consistently match.
When applied to major African cities including Kampala and Maputo, the technology demonstrated more precise slum area classification than existing methods.
The research received the Best Paper Award in the "AI for Social Impact" category at AAAI 2026, the world's most prestigious artificial intelligence conference. Only two papers were selected for this top honor from 693 submissions in the category, confirming that the Korean research team's innovative AI technology has reached the highest global level not only in technical advancement but also in creating tangible social value.
The researchers expect the technology to be applicable across various policy areas, including urban infrastructure planning in developing countries, identification of disaster and disease-vulnerable zones, selection of targets for housing improvement projects, and monitoring progress toward UN Sustainable Development Goals.
"This research demonstrates that AI can contribute to solving social problems even in data-scarce regions, going beyond being a simple analytical tool," Professor Cha said.
KAIST School of Computing researchers Lee Su-min and Park Sung-won participated as co-first authors. The research findings were presented at AAAI 2026 held in Singapore in January.