Internship - 6 months master position in LIG, Grenoble

Contact: Didier Schwab (didier.schwab@imag.fr) and Lorraine Goeuriot
(lorraine.goeuriot@imag.fr)

Description:
Medical data is nowadays generated in a numeric fashion, but is often
poorly structured and difficult to investigate, partiularly when free
text is involved. Building and exploiting patient trajectories from
such heterogeneous and potentially noisy data is a very challenging
task. Artificial intelligence allows to build such medical trajectories
and use them for prediction.

Our work is inspired by the success of several studies that adapt the
neural architecture of (Vaswani et al., 2017) transformers for the
medical domain; including (Lee et al., 2019), (Alsentzer et al., 2019)
or (Huang et al., 2019), all of which have introduced new variants of
the BERT (Devlin et al., 2019) model, pre-trained on general text and
then on medical corpora, and show state-of-the-art results on tasks
such as patient mortality prediction, medical entity recognition,
document classification, etc.

The purpose of this internship is to apply current French transformers
models such as FlauBERT (Le et al., 2020) to new medical use cases. To
do so, the work carried out will follow several steps:

1.  Firstly, a dataset exploitable by neural networks will be created
    from the data gathered in a unit from the Centre Hospitalier
    Universitaire (CHU)

2.  Secondly, existing models built within our team will be applied to
    the new use cases 3. Thirdly, improvement of these models to better
    suit the use cases will be proposed

The intern will be co-supervised by Lorraine Goeuriot and Didier Schwab
from the Laboratory of Informatics of Grenoble (LIG) and Thomas Jouve,
from the Centre Hospitalier Universitaire de Grenoble Alpes (CHUGA).

During the internship, the student will work on pre-training and
evaluation of various models. This requires research and engineering
skills: creation of scripts handling new data, writing and running code
to train transformers, data and evaluation analysis, reading and
understanding of state-of-the-art papers of the domain.

A good level in French would be a plus but is not mandatory. This
project is part of the MIAI Artificial Intelligence and Language chair
(https://miai.univ-grenoble-alpes.fr/research/chairs/perception-interaction/artificial-intelligence-language-850480.kjsp) and de MIAI MyWayToHealth chair
(https://miai.univ-grenoble-alpes.fr/research/chairs/health/my-way-to-health-trajectories-medicine- 851203.kjsp) and as such may lead to a PhD thesis.