[Defended thesis] Benjamin Deneu

[Defended thesis] Benjamin Deneu: Interpretability of distribution models of plant species communities learned through deep learning - application to crop weeds in the context of agro-ecology

Benjamin defended his PhD on 24 November 2022 @ Amphithéâtre Jean-Jacques Moreau, Campus Saint Priest (860 Rue de St - Priest, 34090 Montpellier)

Interpretability of distribution models of plant species communities learned through deep learning - application to crop weeds in the context of agro-ecology

After a BCPST preparatory course, I trained as an Agroparistech engineer, specializing in data science and computer science in my final year, in a double degree with Paris Dauphine University. I did my end-of-study internship in the same team as my PhD, where I started working on similar problems. After 1 year working as a research engineer at INRA and Inria, I wanted to continue working in research by starting a thesis. I'm based in the Inria Zenith team at LIRMM and in the UMR AMAP at CIRAD.
My subject concerns new approaches to species distribution models, based on advances in convolutional neural networks in recent years. These new models have the advantage of being able to use high-dimensional environmental data and capture more complex and richer ecological information than the majority of state-of-the-art models. However, their complexity limits their interpretability. The aim is to study the learning and predictions of these models in order to better interpret them and extract knowledge that could be generalized to other models. The context of the study is the distribution of plant species, and more specifically an application to crop weeds.
Between computer science and agroecology, two fields I'm very fond of, this subject fits in with my multi-disciplinary training and allows me to continue studying these areas as well as using the skills I've acquired on these subjects.

  • Starting date: October 2019
  • University : University of Montpellier
  • PhD school: I2S – Information, Structures, Systèmes
  • Scientific field: IT
  • Thesis management:  Alexis Joly (Inria LIRMM), François Munoz (Université Grenoble Alpes)
  • Thesis management:  Pierre Bonnet (Cirad AMAP), Maximilien Servajean (Université de Montpellier LIRMM)
  • Funding: #DigitAg – Inria
  • #DigitAg : Cofunded thesis – Axe 5 – Challenge 1

Keywords: Deep Neural Networks, Knowledge transfer, Interpretability, Interactions, Landscape, Agricultural practice, Biodiversity, Crop weeds, Agrecology

Résumé : The modelling of interactions between biodiversity, landscape and agricultural practice is one of the major challenges of agro-ecology. Very recently, environmental species distribution models based on deep neural networks have begun to emerge. These first experiments showed that they could have a strong predictive power, potentially much better than the models used traditionally in ecology. One of their advantages is that they can learn an environmental representation space common to a very large number of species so that the prediction performance can be stabilized from one species to another. A first objective of the thesis will be to extend such transfer learning principle to the context of agro-ecology. In particular, data characterizing the landscape and the agricultural practices will be integrated for the prediction of crop weeds and/or associated functional traits. The second objective of the thesis will be to remove the lock on the interpretability of these models in order to deduce new tangible knowledge in agro-ecology. This will include qualifying the environmental representation space learned by the neural network in its terminal layers, typically the last layer of description on which the final linear regression or classification is based. The variables (neurons) in this representation space necessarily correspond to deterministic ecological and environmental patterns, but their exact nature is totally unknown. In the case of the deep agri-environmental models targeted in the thesis, these patterns will also integrate information from the landscape and agricultural practices. Their analysis will provide a better understanding of whether or not massive integrative approaches, based on a wide variety of input data, are needed, or whether they should focus on certain key factors.

Jury compound:

  • Christine MEYNARD, Chargée de recherche, INRAE - Rapportrice 
  • Christophe RANDIN, Privat-docent, Université de Lausanne - Rapporteur 
  • Laure BERTI-ÉQUILLE, Directrice de recherche, IRD - Examinatrice
  • Alexis JOLY, Directeur de recherche, Inria - Directeur de thèse
  • François MUNOZ, Professeur, Université Grenoble Alpes - Directeur de thèse
  • Pierre BONNET, Chargé de recherche, Cirad - Encadrant
  • Maximilien SERVAJEAN, Maître de conférence, Université Paul-Valéry Montpellier - Encadrant

Contact  benjamin.deneu [AT] inria.fr​ – Tel : +33 (0)

Social networks: ResearchGateLinkedIn


See also


Modification date: 23 August 2023 | Publication date: 19 August 2022 | By: ZM