The AFRIF 2021 thesis prize awarded to Ignacio Rocco

The AFRIF (Association Française pour la Reconnaissance et l'Interprétation des Formes) awards each year its thesis prize. The objective of this award is to highlight and encourage the best doctoral work on these topics and to boost research in the field. The AFRIF 2021 thesis prize was awarded to Ignacio Rocco and a special mention was made to Hugo Richard. AFRIF recently presented the winner of its 2021 thesis award. The finalists had to have defended their thesis in the fields of image recognition between July 1st, 2020 and December 31st, 2021 in a French doctoral school or in a cotutelle with a French doctoral school. The theses were evaluated by a jury chaired by Vincent Lepetit (ENPC, Paris) and composed of members of the Board of Directors including Marie-Odile Berger (INRIA Nancy - Grand Est), President of AFRIF.

AFRIF 2021 thesis prize awarded to Ignacio Rocco

The AFRIF 2021 thesis prize was awarded to Ignacio Rocco for his work entitled " Neural Architectures for Estimating Correspondences Between Images ", carried out at the Ecole Normale Supérieure - Université PSL and under the supervision of Josef Sivic and Relja Arandjelović. The thesis focuses on the development of methods for matching between pairs of images in challenging situations such as extreme lighting change, scenes with little texture or including repetitive structures or matching between parts of objects that belong to the same class but may have large intra-class differences in appearance. Ignacio Rocco contributes as follows:

1. Develop a trainable approach for parametric image alignment using a Siamese network model
2. Design a weakly supervised training approach that allows training from real image pairs annotated only at the image pair level
3. Propose Neighborhood Consensus Networks that can be used to robustly estimate matches for tasks where discrete matches are required
4. Develop a more efficient variant capable of reducing the memory requirements and execution time of Neighborhood Consensus Networks by a factor of ten

Special mention to Hugo Richard

A special mention was also given to Hugo Richard for his work entitled " Unsupervised Component Analysis for Neuroimaging Data ", done at the University of Paris-Saclay under the supervision of Bertrand Thirion. This work is a computer science and mathematics thesis that applies to the field of neuroscience and more particularly to research on the modeling of human brain activity by electrophysiology and imaging. The need to use mathematical tools arises from the complexity of identifying neural activity from data collected from so-called natural stimuli. In this thesis, the author first considers the case of the shared response model, in which subjects are assumed to share a common response. It is useful to reduce the dimension of the data but its training remains expensive for functional imaging data whose dimension can be immense. This model also makes unrealistic assumptions about the above data. Another method that makes more realistic assumptions is independent component analysis, but it is difficult to generalize to datasets containing multiple subjects. Hugo Richard therefore proposes an extension of the latter which he calls multiview ICA, it is based on the maximum likelihood principle which is suitable for multisubject datasets. The first AFRIF prize of the 2021 edition has been rewarded with 1 000€ and will be invited to present his work during the RFIAP 2022 young researchers conference which will be held in Vannes from July 5 to 8, 2022. Translated from Le prix de thèse AFRIF 2021 décerné à Ignacio Rocco