Offre en lien avec l’Action/le Réseau : – — –/– — –
Laboratoire/Entreprise : ICube, University of Strasbourg
Durée : 3 years
Contact : lampert@unistra.fr
Date limite de publication : 20/3/2020
Contexte :
The adoption of deep learning techniques in medical imaging applications has been limited by the availability of the large labelled datasets required for robust training, as well as the difficulty of explaining their decisions. This thesis will make contributions towards overcoming both of these limitations.
Sujet :
It will achieve this by developing approaches to learn more robust representations using explainability. These approaches will be referred to as self-regularised deep learning in the presence of limited data. The problem of domain adaptation and learning domain invariant representations in histopathological whole slide segmentation will be taken as the initial focus of this study, but this is open be expanded during the project. Current approaches fail to achieve domain invariance because of the large domain shifts between histochemical and immunohistochemistry stainings.
An initial research direction will be to develop novel training mechanisms that are aware of, and therefore avoid, situations in which the network focusses only on limited parts of the salient information (as defined by the expert through few manual annotations) will be developed. These will force a more general representation to be learnt. The benefit being threefold: the model will be more generalisable, more domain invariant, and more amenable to transfer learning.
Profil du candidat :
The position is open to both foreign and French students who hold a Master’s degree in Computer Science. French is not necessary, but the candidate must be confident in spoken and written English.
Formation et compétences requises :
The candidate must have a good mathematical background, skills in machine learning (supervised and/or unsupervised). Experience in deep learning and representation learning would be a plus.
Adresse d’emploi :
ICube UMR 7357 – Laboratoire des sciences de l’ingénieur, de l’informatique et de l’imagerie
300 bd Sébastien Brant – CS 10413 – F-67412 Illkirch Cedex
Document attaché : 202004301542_PhD_advert.pdf