Central and East Africa Center

The aim of the first line of research is to develop and analyze the dynamics of mathematical and computer models that model the spread of infectious diseases, as well as their control and prevention.
The aim is to support public health decisions for better upstream design and downstream monitoring of health programs implemented in Central Africa, and more specifically in Cameroon. The main objective of the second line of research is to study the epidemiology and management of tropical crop diseases in relation to climate change (Kolaye et al. 2018, Fotso et al. 2022, Dountio et al. 2024). We are interested in mirid pests of cocoa, nematodes of plantain, coffee bark beetles, paracoccus marginatus pests of papaya, busseola fusca pests of maize, dysmicoccus brevipes pests of pineapple and phenacoccus manihoti, pest of cassava.
The One Health modeling approach will use the close and interdependent links between human, animal and environmental sciences to create new mathematical models and methods, taking into account changes in human and animal living environments that favor the emergence or (re)emergence of new pathologies. These models and methods will help in epidemiological surveillance, disease control and prevention. This project will be carried out in collaboration with the Programme National de Prévention et de Lutte contre les Zoonoses Émergentes et Ré-émergentes (PNPLZER), the Centre Pasteur du Cameroun, the Ministry of Public Health and the Cameroon National Meteorological Office.

The first area of research is the development of automatic speech recognition models for African languages. African languages are mostly spoken rather than written.
However, many of these languages have writing systems. Both text and audio data are very scarce for these languages, making it difficult to effectively exploit the most powerful AI models.
In addition, African languages have linguistic characteristics that are very different from the languages most commonly used in AI tools. The aim of this research work is to develop specific models (exploiting transfer learning and multilingual models) for converting speech into text, which, among other objectives, will make it possible to create more text corpora, and enable the development of artificial learning models for machine translation from African languages into French and English and vice versa. As in the case of speech, the center will focus on the challenges of low data quantity, data quality, low levels of annotated or aligned data, and linguistic characteristics.
The second line of research is the development of deep learning models for image-based flood monitoring in urban areas. Climate change and socio-economic factors such as urbanization are causing environmental disasters in urban areas in Central Africa. More and more floods are occurring suddenly, affecting previously unscathed areas. The center will work on deploying drones to capture images of areas of interest and develop deep learning models for flood prediction.
In addition to overcoming the lack of data, the work aims to couple the mathematical models of fluid diffusion with those of deep learning.
Finally, a scientific question common to all the topics addressed in this theme is that of the explicability of learning models. Two levels of explicability will be targeted in the work of the center's researchers: the level of professional experts (linguists, hydrologists, etc.) and the level of political decision-makers (ministry in charge of primary education for training in African languages, mayors of urban communities for flood management).

Complementary to the previous one, this line of research aims to develop tools to collect the data needed for the models and produce new knowledge for users. These data will be used by machine learning algorithms (deep learning in particular) to build models for speech recognition and automatic translation of texts in African languages. The first scientific issue is the formalization of data collection processes for environments that are not very digitized, with sensors that need to repatriate data collected dozens of kilometers away, and actors from different horizons and disciplinary fields (linguists, anthropologists, rural populations, etc.) that need to be mobilized.
In addition to text and audio data, the center will also work on collecting environmental data in the form of images and videos from drones. This data is essential for the deep learning algorithms used to monitor flooding in urban areas. The second research question, common to both applications, is that of developing embedded systems to host the pre-processing models developed at the center. The third scientific question is that of scaling up data collection through technology (chatbots, voicebots, drones) and with the involvement of target citizens. Involving citizens in data collection requires the deployment of training and information resources on the issues, opportunities and techniques involved. Finally, the fourth scientific issue is that of data storage and dissemination, with a view to facilitating their integration into learning models.