DO Salomé
École normale supérieure - PSL
salome.do [at] ens.psl.eu
Short bio
MSc / Engineering degree at ENSAE IP Paris
Thesis topic
Computational Content Analysis Methods for News Frames Prevalence Estimation in the Political Press.
Short abstract
This dissertation aims at providing Computational Content Analysis (CCA) methods for the analysis of News Framing in the political press. First, it aims at creating a french corpus of political press articles and providing human annotations for two news frames identification tasks, derived from the literature on strategic news framing and “horse race” journalism. Second, it aims at exploring the modalities (frame complexity, data quantity and data quality) in which Supervised Machine Learning (SML) methods can “augment” social scientists, i.e. train a model to generalize social scientists’ content analysis (CA) codebook (and subsequent text annotations) so that billions of articles can be analyzed instead of a few hundred. Third, the dissertation aims at evaluating the potential benefits of CCA over CA when it comes to estimating news frames prevalences in a corpus. What justifies using CCA over CA, and is it always justified? I will try to define the conditions on SML models performances under which news frames prevalence estimates are more accurate with CCA than CA.
Short bio
Master’s degree in Theoretical Physics, 2021, Technical University Berlin
Thesis title
Analysing discourse and semantics through geometric representations.
Short abstract
I explore geometric approaches to language and discourse analysis. Currently, I work on combining methods from network science and natural language processing to gain insights on the mechanisms behind information and knowledge spreading related to climate change.
VINCENT Louis
Université de Paris / Inria / Inserm / Implicity (CIFRE thesis)
louis.vincent [at] implicity.fr
Short bio
Master 2 – Mathématiques, Vision & Apprentissage (ENS Paris-Saclay),
Master 2 – Statistiques (Sorbonne Université – Campus Pierre et Marie Curie)
Thesis title
Longitudinal data encoding applied to medical decision support in telecardiology.
Short abstract
In telecardiology as in many other fields of modern medicine, we have at our disposal large amounts of data explaining the evolution of a patient. These data can often be missing or corrupted, and data from several sources can sometimes be of different nature, which makes their exploitation difficult.
My goal is to develop a model capable of synthesizing different types of temporal data via auto-encoders to infer the state of a patient. In the context of tele-cardiology, this could for instance allow us to predict deteriorations of a patient’s health status, and thus anticipate and prevent more serious complications.
DELEMAZURE Théo
Université Paris Dauphine-PSL
theo.delemazure [at] dauphine.eu
Short bio
Master 2 IASD (ENS Ulm)
Thesis title
Context-dependent collective decisions.
Short abstract
Traditional research in social choice (axiomatic or computational) consists in proposing collective decision mechanisms that are often too rigid. We will focus on the development of families of parameterized rules, which have enough variability and flexibility to allow the development of “tailor-made” mechanisms for specific problems.
ARJONILLA Jérôme
jerome.arjonilla [at] dauphine.psl.eu
Short bio
Master in Computer Science, Sorbonne University + Double Bachelor in Mathematics and Economics, Université Toulouse
Thesis title
Search and Learning algorithm for games with imperfect information.
Short abstract
Monte Carlo research has revolutionised game solving, and, combined with deep reinforcement learning has led to the creation of programs such as AlphaGo, Alpha Zero or Polygames that have beaten the best human players in many games. My thesis topic will focus on the extension of the methods of Monte Carlo and Deep Reinforcement Learning for imperfect information games with several players (multi-agent system).
BERNARD Theo
Université Paris Dauphine-PSL
tbertrand [at] ceremade.dauphine.fr
Short bio
Diplôme d’ingénieur et Master en Mathématiques appliquées de l’Ecole Centrale Lyon
Thesis title
Méthodes géodésiques et IA pour la microscopie par localisation ultrasonore.
Short abstract
Ultrasonor Localization Microscopy is a new method in super-resolved Medical Imaging that allow us to overcome compromise between precision and penetration distance in the tissues for the imaging of the vascular network. This new type of images raises new mathematical questions, especially for the segmentaton and analysis, necessary steps to achieve medical diagnostic of patients.
MAKAROFF Nicolas
Université Paris Dauphine-PSL
makaroff [at] ceremade.dauphine.fr
Short bio
M2 MVA ENS Paris-Saclay
Thesis title
Segmentation and modeling of tree structure by Deep Learning with geometric constraints, applications in biomedical imaging.
Short abstract
Deep Learning has shown real capabilities for different tasks ranging from classification to segmentation in various fields such as chemistry, computer vision or even medicine. Generally, known architectures are used to solve these problems. However, this generality of architectures, although obtaining good results, does not yet consider the geometric and topological structure of the studied data, which can lead to a reduced interpretability and acceptability of the results due to a lack of transparency
EVEN Mathieu
mathieu.even [at] inria.fr
Short bio
M2 Orsay/Paris-Saclay
Thesis title
On Federated and Distributed Learning Problems.
Short abstract
We study Federated and Distributed learning problems, with a strong incentive on theoretical guarantees. A (possibly large) number of agents aim at making predictions (supervised or unsupervised learning). To what extent can they benefit from collaborating with each other, depending on the learning problem and the communication constraints ?
Short bio
Ingénieur polytechnicien, École polytechnique (Palaiseau)
MPhil in Computational biology, University of Cambridge (UK)
Thesis title
Combine artificial intelligence with high resolution microscopy to better dissect the mechanism of binding and mechanism of action of multi-specific biologics.
Short abstract
Photo-activated localization microscopy (PALM) enables high-resolution recording of single proteins trajectories in live cells, thus providing precious probes of small-scale properties of the protein environment. I use graph neural networks to characterize relevant physical properties of these dynamics, and developed a flexible analysis scheme able to deal both with the diversity of motion types encountered in nature and with the fact that observed trajectories inevitably differ, to some extent, from archetypal theoretic models.
TRIBOULIN Amaury
amaury.triboulin [at] inria.fr
Short bio
Master’s Degree at Ecole normale supérieure
Thesis title
Symmetries in Machine Learning for Structured Data.
Short abstract
In this thesis, we will consider high-dimensional problems with an additional structure that comes from the geometry of the input signal and explore ways to incorporate this geometric structure into the learning algorithms. We have already started to investigate new architectures based on equivariant layers which we tested on combinatorial optimization problems and showed that it is possible learn representations of hard (typically NP-hard) problems. We believe this could lead to new algorithms, less resource-dependant, for learning efficient heuristics for practical instances.
