Embryonic development is a complex hydrodynamic problem. Global flows stem from individual cell-cell rearrangements, which can either resist or drive embryonic flows. Tracking their location and orientation is crucial to understanding morphogenesis, yet it remains a computationally challenging task. Here, in this M1/M2 internship, we propose to adapt a method that we have proven works for detecting cell divisions. This will be the starting point for a theoretical PhD. Indeed, preliminary results in our group support that modulation in the rates of cell-cell rearrangements triggers phase separation between cell types, but we seek to develop simplified analytical models to account for this effect. Please also note that: (1) We have M2 and PhD internship openings in our group to explore this subject, but we could also recruit a motivated M1 student. (2) We will work in close collaboration with the groups of Sham Tlili, Virgile Viasnoff, Pierre-François Leone & Thomas Lecuit. (3) We do not require any previous experience in deep learning nor in biology ; this subject will be an opportunity to learn tools such as CNN or GAN, and I have an experience in supervising physics students on these subjects; the project is interdisciplinary, but addresses physics questions; (4) Marseille Luminy is arguably the most beautiful campus in the world, right in the middle of the Calanques National Parc.