Annonce

 

The context of the study is an ANR project [1] which proposes to analyze, through the prism of three major crises (asylum, Covid-19, war in Ukraine), the interactions between international migration and the associated inequalities. The notion of a migratory journey implies the idea of continuity, without being synonymous with linearity or a predefined direction. Migrants share itineraries marked out by places and evolve within one or more collectives, notably through images. These images, made up of objects (tents, vehicles, everyday objects, etc.) and references (logos, landscapes, etc.) referred to under the generic term of "signs", enable to reconstruct a reality, that of the migrant's daily life, to be compared and mirrored with official discourse and images. The proposed approach is designed to :

- identify these signs in images: this will involve the use of object detection and recognition techniques, with a focus on deep learning approaches. A state-of-the-art review of classic methods (R-CNN family, Panoptic Feature Pyramid Networks, PointRend, OneFormer, SAM, etc.) will enable to identify, depending on the context, an efficient algorithm for detecting objects of interest,

- integrate expert knowledge, using first-order logic, into a tensor formalism derived from LTNs,

- train this neurosymbolic approach to determine links between signs,

- deduce a semantic interpretation to "tell" the migrant's story.

 

This story will then be compared and mirrored to official speeches and images.