%0 Conference Proceedings
%9 ACTI : Communications avec actes dans un congrès international
%A Ben Abbes, A.
%A Machicao, J.
%A Meneguzzi, L.
%A Corrêa, P.L.P.
%A Specht, A.
%A David, R.
%A Subsol, G.
%A Vellenich, D.F.
%A Devillers, Rodolphe
%A Stall, S.
%A Mouquet, N.
%A Chaumont, L.
%A Berti-Equille, Laure
%A Mouillot, D.
%T Checklist strategies to improve the reproducibility of Deep Learning experiments with an illustration [poster]
%D 2022
%L fdi:010087182
%G ENG
%I 
%P 1  multigr.
%R 10.5281/zenodo.6587702
%U https://www.documentation.ird.fr/hor/fdi:010087182
%> https://horizon.documentation.ird.fr/exl-doc/pleins_textes/2023-07/010087182.pdf
%W Horizon (IRD)
%X The challenges of Reproducibility and Replicability (R&R) have become a focus of attention in order to promote open and accessible research. Therefore,  efforts have been made to develop good practices for R&R in the area of computer science. Nevertheless,  Deep Learning (DL) based experiments remain difficult to reproduce by others due to the complexity of these techniques. In addition, several challenges concern the use of massive and heterogeneous data that contribute to the complexity of this R&R. Firstly, we compiled three different aspects to help researchers to improve R&R. This compilation was based on machine learning checklists, guidelines, and principles from FAIR. Therefore, this compilation is useful for a (1) researcher seeking to reproduce a paper, (2) an author reporting on an experiment, and (3) a reviewer seeking to qualify the scientific contributions of the work. Secondly, we illustrate the compilation of three recent DL experiments for socio-economic estimation using remotely sensed data.
%B RDA.Research Data Alliance
%8 2022/06/20-23
%$ 122 ; 126 ; 021