PhD seminars – 2021/2022

PhD seminar – 27th, june 2022

Buntheng Ly – Université Côte d’Azur

Title: Exploring The Relationship Between Left Ventricular Wall Thinning And Post-infarction Ventricular Arrhythmia Using Explainable Deep Learning On Computed Tomography Images

Abstract: The sudden cardiac death (SCD) is a major health concern in the developed country accounted for up to 400 000 deaths annually in Europe, as well as the United State. The studies have shown that ventricular arrhythmia (VA) is the phenomenon most frequently occurred leading up to the SCD.

In this project, we proposed a fast and fully-automated prediction of the VA using the non-invasive cardiac imaging modality, the cardiac CT. We built our algorithm based on the left ventricular wall thinning, an indicator of the myocardial infarction scar, which is the main anatomical substrate in the arrhythmia mechanism. Using the Deep Learning model, our pipeline can produce, in less then 3 mn, a VA prediction with 79.2% accuracy, compared to the 67.1% accuracy of the current clinical marker of VA (left ventricular ejection fraction).

 


Melissa Antonelli – Université de Bologne

Titre: Randomized proofs-as-programs

Résumé: La correspondance dite de Curry-Howard définit l’une des interactions les plus intrigantes entre la logique et la théorie des langages de programmation.

Melissa Antonelli

Dans sa forme minimale, elle relie la logique propositionnelle intuitionniste et le λ-calcul simplement typé, mais, au cours des dernières décennies, elle a été généralisée de plusieurs façons. Je présente un travail conjoint avec U. Dal Lago et P. Pistone, dont la principale contribution consiste à étendre ce résultat au cadre probabiliste. Nous fournissons une correspondance entre une version intuitionniste de la logique propositionnelle de comptage, et un système de type expressif pour le λ-calcul d’événements probabilistes, dans lequel l’évaluation CbV et CbN de programmes fonctionnels aléatoires discrets peut être simulée. Dans ce contexte, les preuves (typage) ne garantissent pas la validité (terminaison), mais révèlent les probabilités sous-jacentes. terminaison), mais révèlent la probabilité sous-jacente.

PhD seminar – 13th, june 2022

Florent Jousse – team EPIOME, Inria Sophia-Antipolis

Title: How to train face morphable models with non-intersecting subspaces?

Abstract: This presentation shows how to learn deformation spaces for 3d face meshes. Typically, a first deformation subspace encodes the possible identity changes while another encodes the possible facial expressions. It has been proposed to jointly use two Principal Component Analyses (PCA) to find these subspaces. PCA chooses the basis vectors pointing in the direction of maximum variance when the input shapes are transformed in these low dimensional subspaces. However, PCA produces an intersection between the subspaces for expressions and identities. The problem with this intersection is that modifying facial expressions will also modify the person’s identity and limit such models’ usefulness. Therefore, we formulate an optimization problem penalizing subspaces’ orthogonality. Then, we propose an iterative procedure to solve this problem. Our approach ensures a quasi-orthogonality between the subspaces and improved facial attributes and expressions manipulations of existing models.

 

Les commentaires sont clos.