Know more

Mixing machine learning and semantic web methods for the optimization and planning of market gardening in agro-ecology

I'm a Cifre thesis student in the Web3 team at the Montpellier Laboratory of Computer Science, Robotics and Microelectronics (LIRMM) and the Elzeard company.
My background is in computer science. After my bachelor's degree in computer science at the Faculty of Science in Montpellier, I continued within the same structure on the DECOL master's degree, focusing on Data, Knowledge and Natural Language. I did a research internship at Lirmm on the theme of neural networks. Following my research experience, I preferred to go into industry, but with a well-developed taste for research, I turned to a CIFRE thesis.
The idea of my thesis topic is to develop machine learning algorithms using data from knowledge graphs for tasks such as crop selection or rotation planning. The agricultural domain requires taking into account agronomic, contextual, meteorological and other parameters. My aim is to take into account most of these parameters and produce algorithms that can manage all of them to make personalized recommendations to farmers.

I believe that agriculture is a key sector in the functioning of our societies. Farmers have a difficult job, and I think it's important to try and make their lives easier by saving them time on organization and providing them with a wealth of information that will benefit them. What's more, putting my knowledge and skills at the service of ecology by trying to develop agro-ecology techniques among producers and make them more accessible is, for me, interesting.

Agriculture is a field that requires both theoretical knowledge and experience. With larger and larger data sets collected from growers, and aggregated and formalized business knowledge, it would be possible to develop increasingly advanced planning algorithms, taking into account a multitude of parameters such as soil health, climatic hazards, the appearance of pests and so on.

• Starting date : 2nd February 2021
• University : Université de Montpellier
• PhD school : I2S – Information, Structures, Systèmes
• Scientific fiels : IT
• Thesis management : Clément Jonquet (Lirmm, Université de Montpellier)
• Thesis supervisors: Konstantin Todorov (Lirmm, Université de Montpellier), Florence Armadeilh (Elzeard)
• Funding: Convention Cifre
• #DigitAg : Labeled PhD – Challenge 1 : Le challenge agroécologique

Mots-clés : Information - Market gardening - Agroecology - Data - Semantics

Résumé : L’objecif général de la thèse est de produire une méthodologie de traitement des données agricoles pour les enrichir sémanquement (via leur description ou leur annotation avec des ontologies ou des référentiels du domaine (coproduits dans le contexte de D2KAB)), puis les interconnecter et les désambiguïser afin de produire un graphe de connaissances à exploiter via les algorithmes de plongement de graphes (graph embeddings) qui sont une méthode d’apprentissage de représentation. Nous nous intéresserons tout particulièrement aux recommandations en termes de planification des cultures, de localistion des cultures et de compagnonnage, c’est-à-dire avec quoi planter. Nous faisons l’hypothèse que l’enrichissement sémantique des données agricoles augmentera la performance des méthodes d’apprentissage automatique pour la tâche de recommandation que l’exploitation des données brutes.

Contact : darnala.b@gmail.com - Tel: 06.58.91.23.94