Explainable AI and poverty prediction
| Title | Explainable AI and poverty prediction |
|---|---|
| Summary | Provide explanations of AI data-driven poverty predictions in sub-saharan africa |
| Keywords | Satellite data, transfer learning, explainable AI, Africa, economic development indicators |
| TimeFrame | Spring 2022 (but welcome to start sooner) |
| References | (1) Lee and Braithwaite (2021), "High-Resolution Poverty Maps in Sub-Saharan Africa", https://arxiv.org/abs/2009.00544
(2) Jean, Burke, Xie, Davis, Lobell, and Ermon (2016), "Combining satellite imagery and machine learning to predict poverty", Science, https://www.science.org/doi/10.1126/science.aaf7894 (3) Roscher, Bohn, Duarte, and Garcke (2020), "Explainable Machine Learning for Scientific Insights and Discoveries", IEEE Access, https://ieeexplore.ieee.org/document/9007737 |
| Prerequisites | Machine learning, deep learning |
| Author | |
| Supervisor | Thorsteinn Rögnvaldsson, Mattias Ohlsson |
| Level | Master |
| Status | Open |
The task is to first reproduce some of the results in Lee & Braithwaite (2021), then build further on that to explain what features/properties that are important for the predictions and also how. Data is available.
This is a software project involving high volumes of data. The implementation of the model is not trivial and requires very good programming skills and understanding of machine learning tools. The work is done in cooperation with the Department of Human Geography, Lund University.
WP 1: Literature study to understand and define the problem. WP 2: Reproduce the model in Lee & Braithwaite (2021) for a few countries. WP 3: Choose a few methods for feature importance analysis and "XAI" and apply them to this model.
Ideal outcome: a study of a current and important topic that will form the basis for a scientific contribution.
