Defense of Théophile Ghislain Loïc Eyango Tabi thesis

Membres du jury :

• Reviewers before defense :
  • Christine LARGOUET, Senior lecturer, Agro Institut Rennes-Angers, IRISA, Rennes, France
  • Paul HONEINE, Professor, University of Rouen, LITIS, Rouen, France
• President : Catherine BELLOC, Professor, Oniris, Nantes, France
• Examiners :
  • Ludovic BROSSARD, Research engineer, INRAE, PEGASE, Rennes, France
  • Christine FOURICHON, Professor, Oniris, Nantes, France
  • Christine LARGOUET, Senior lecturer, Agro Institut Rennes-Angers, IRISA, Rennes, France
  • Paul HONEINE, Professor, University of Rouen, LITIS, Rouen, France
• PhD director:
  • Sébastien PICAULT, Research fellow, INRAE, BIOEPAR, Nantes, France
• PhD co-director:
  • Nicolas PARISEY, Research engineer, INRAE, IGEPP,
• Industrial supervisor :
  • Xavier L'HOSTIS, Innovation manager, Adventiel, Rennes, France
  • Victoria POTDEVIN, Data science manager, Adventiel, Rennes, France

Résumé :

Effective management of infectious diseases in livestock requires detecting and forecasting outbreaks despite the complexity of host–pathogen–environment interactions and the difficulty of extracting relevant information from farm sensors. This thesis proposes an innovative approach that directly couples sensor data with knowledge derived from mechanistic epidemiological modeling. By combining deep learning, which can automatically extract patterns from complex signals, with mechanistic models based on veterinary expertise, we aim to improve both short-term diagnosis and longterm disease predictions in livestock. The main contributions of this work are: (1) coupling empirical sensor data with mechanistic simulations to fill the gap between sensor-based observations and theoretical knowledge;
(2) explicitly incorporating uncertainty into predictions to enhance reliability; and (3) developing a method to differentiate pathogen-specific scenarios to guide targeted interventions. Applied to respiratory diseases in young cattle (BRD), our
methods have demonstrated, under both real and simulated conditions, their ability to automate shortterm diagnosis and long-term predictions BRD dynamics, thereby significantly reducing antibiotic use and improving farm performance. This work opens new perspectives by proposing a modular methodology that combines sensor data and knowledge, potentially serving as an innovative decision-support tool for optimized health management.

Mots clés :

Sensors ; Artificial intelligence ; Deep learning, Epidemiological mechanistic modelling ; Animal health ; BRD