Research

Speech and language technology for Ivorian languages

Côte d’Ivoire has dozens of national languages, and speech recognisers and language models barely cover them. My aim is to build the data and the tools that are missing, starting with Dioula, and to train Ivorian students to carry this work forward.

Current work covers corpus design (read and spontaneous speech, dialect coverage, speaker balance), collection and validation tooling, automatic pre-transcription with multilingual models such as MMS, and fine-tuning of speech recognition models once the corpus is complete. See MEY-D26 and KAN.

AI for health

With physicians from the Faculty of Medicine of UFHB, I study how AI can support care in African hospitals. This includes a multi-agent simulation of clinical management in Mesa that integrates a fuzzy patient satisfaction model, interpretable machine learning to predict thrombocytopenia in patients with atrial fibrillation, and survey-based models of public attitudes towards tuberculosis patients.

This work relies on real African data, including the AFRICA registry of atrial fibrillation, and favours explainable models that a clinician can discuss: prediction of mortality, of tuberculosis knowledge and of return-to-work decisions after treatment.

Urban mobility

Abidjan has few traffic sensors. Using trajectory data from an app-based mobility platform, my colleagues and I showed that supervised models, Random Forest in particular, can estimate traffic density on about 800 road segments, and that speed limits and road hierarchy matter more than fine geometric detail. Traffic modelling continues in a doctoral thesis I supervise.

Energy and smart meters

With Dr Yazid Hambally Yacouba, I designed a multi-agent system to collect and schedule data from smart electricity meters, then aggregation methods based on the jump process to analyse consumption at the scale of a home, a building or a neighbourhood.

Agriculture

With ANADER, we build image datasets and deep learning models to recognise diseases of yam, cassava and plantain, crops that feed much of the country but get little attention in the computer vision literature.

Multi-agent systems and decision support

My earlier work, carried out in Morocco, used multi-agent systems, ontologies, fuzzy logic and AHP to make public procurement more transparent. The same architecture now serves other domains: hospital resource allocation and modular malware detection.

Cybersecurity

With Dr Yazid Hambally Yacouba, I work on the detection of malicious PDF files, combining supervised learning with synthetic data generated by CTGAN to deal with class imbalance.

Collaborations

• Prof. Abdellah Azmani, Faculty of Sciences and Technologies of Tangier, Université Abdelmalek Essaâdi, Morocco.
• Prof. Adama Coulibaly, Faculty of Mathematics and Computer Science, UFHB.
• Dr Awa Fofana, Faculty of Medicine, UFHB.
• Dr Yazid Hambally Yacouba, University of Bondoukou.
• Intelligent Automation & BioMed Genomics Laboratory (IABL), FST Tangier, headed by Prof. Abdellah Azmani, since 2023.
• Dr Dethie Dione, Université Gaston Berger, Saint-Louis, Senegal.
• Prof. Eric Matzner-Lober, ENSAE-ENSAI, Paris, France.
• Prof. Abdelghani Chibani, LISSI, Université Paris-Est Créteil, France.
• Masakhane, the African NLP research community.