Presentation

Team presentation

The EVERGREEN team actively works on the design and implementation of cutting-edge machine learning techniques to effectively exploit heterogeneous and multi-temporal Earth observation data for numerous downstream tasks, including land cover mapping, land use following deforestation monitoring, forest variables estimation, and yield prediction to mention a few.
These endeavors directly address modern agro-environmental challenges with the goal to provide tools and insights for a more sustainable exploitation of natural resources.
To this end, the team delves into fundamental research questions related to the transferability of multi-modal classification models, the design of machine learning models for low-data regime scenarios, the interplay between model-centric and data-centric Machine Learning and the interpretability and explainability of classification algorithms for both image and time series data. This aspect is closely tied to the imperative of make the black box models gray, particularly within interdisciplinary research collaborations, such as the ones in which the EVERGREEN team operates daily.

Research themes

The team has three main scientific objectives:

  • Design and propose appropriate Machine Learning methods especially tailored for the specificity of Earth Observation data.
  • Adoption and development of new learning paradigms to support Earth Observation data analysis
  • Ameliorate the interaction between domain experts and machine learning systems following two different paths: i) introduce knowledge-based apriori to guide the learning process and ii) design models that provide explainabilty/interpretability on how the decision is made. 

International and industrial relations

Relations internationales : EPFL, Wageningen University & Research, Univ. Calabria, Univ. Torino, Univ. Parthenope, DLR

 

Relations industrielles : ATOS, CNES, Eco-Med

Comments are closed.