KAIST

<(From Left) Sumin Lee, Sungwon Park, Prof. Jihee Kim, Prof. Meeyoung Cha, Prof. Jeasurk Yang>

"Cities don't even know where their slums (impoverished areas) are located."

In many developing nations, the most vulnerable citizens are invisible to the state simply because their homes don't appear on any official map. Today, a breakthrough using Artificial Intelligence (AI) is changing that.

A joint research team from KAIST and Chonnam National University in South Korea and MPI-SP in Germany has developed an AI technology that autonomously identifies slum areas using nothing but satellite imagery. This technology is expected to fundamentally transform urban policy-making and public resource allocation in developing countries where data is scarce and has won the Best Paper Award in the ‘AI for Social Impact’ category at the AAAI 2026 (Association for the Advancement of Artificial Intelligence), the world's premiermost prestigious AI academic conference.

Why it Matters

While previous studies struggle to recognize slums across countries due to varying architectural styles, the team introduced a "Mixture-of-Experts (MoE)" structure. In this system, multiple AI models learn different regional characteristics; when a new city is inputted, the system automatically selects the most appropriate model.

<Figure1. Overview of the Mixture-of-Experts(MoE) structure to identify slum areas>

The core of this research is "Test-Time Adaptation (TTA)" technology. Even if humans do not pre-mark slum locations in a new city, the AI reduces its own errors by comparing and verifying the prediction results of multiple models, trusting only the areas where they commonly agree. This ensures stable performance even in regions with insufficient data.

The research team applied this technology to major cities such as Kampala (Uganda) and Maputo (Mozambique) and confirmed that it distinguishes slum areas more precisely than existing state-of-the-art technologies.

This technology is expected to be utilized in various policy fields, including:

• Establishing urban infrastructure expansion plans for developing countries.

• Identifying areas vulnerable to disasters and infectious diseases in advance.

• Selecting targets for housing environment improvement projects.

• Monitoring the implementation of UN Sustainable Development Goals (SDGs).

<Figure2.  Slum segmentation results in Kampala in 2015 (yellow) and 2023 (red). Over the eight-year period, the slum ratio in the city increased from 8.4% to 8.6%>

Meeyoung Cha, an AI researcher and author, stated, "This research proves that AI is no longer just a tool for analysis. It is a tool for action. Our technology can bridge the data gap to solve the world’s most pressing social challenges." Jihee Kim, an economist and author, added, "It will complement costly field surveys and help effectively allocate limited resources to the areas that need them most."

The research results were presented at AAAI 2026 in Singapore on January 25th.

• Paper Title: Generalizable Slum Detection from Satellite Imagery with Mixture-of-Experts

• Paper Link: https://aaai.org/about-aaai/aaai-awards/aaai-conference-paper-awards-and-recognition/

This research was supported by the National Research Foundation of Korea (NRF) through the Mid-career Researcher Support Program and the Data Science Convergence Human Resources Training Program.