Created by W.Langdon from gp-bibliography.bib Revision:1.9039
https://cstb.icube.unistra.fr/index.php/Lalla_Aicha_Kon%C3%A9",
https://publication-theses.unistra.fr/public/theses_doctorat/2024/kone_lalla_2024_ED269.pdf",
The first challenge of this thesis was to test the potential of AI for prediction and decision-making based on short time series, in the absence of massive data, which is typically lacking for such purposes. Indeed, the evolution of an annual indicator over 100 years provides just 100 data points, but even this is not available in developing countries.
The first contribution concerns the prediction of baccalaureate admission statistics in Mauritania. As historical data are limited, it is essential to have a prediction technique adapted to short time series. This thesis proposes a tool based on genetic programming and Kalman filter, improving the accuracy of short-term predictions. Validation on various data sets demonstrates the tool's superiority over literature techniques, promising significant improvements in the Mauritanian education system.
On the other hand, given the risks and threats it poses, the use of AI raises questions and calls for ethical principles to be taken into account, which must necessarily be aligned with local socio-cultural norms; another major challenge.
The second contribution presents an intuitive solution using generative AI to enable governments to establish local ethical principles for AI software. This solution consists of two web applications: one for governments and the other for software developers. The government-focused application dynamically calibrates ethical weights in different domains according to socio-cultural context, creating tailor-made ethical plans for each domain. As for the developer application, it actively assesses the compliance of software with the ethical principles established by the government and provides feedback for their possible recalibration.
Together, these contributions pave the way for the ethical deployment of AI in line with local values, and for improved forecasting for public service improvement in developing countries",
ED 269 Supervisor Pierre Collet",
Genetic Programming entries for Lalla Aicha Kone