Supervision

Postdoctorate

  • 2015: Muhammad Muzzamil Luqman - SuperClass: new methods of acquisition and classification for high content screening of membrane receptor organization and dynamics using super-resolution microscopy - with Macha Nikolski

Official PhD Thesis supervision

  • 2019-2022: Adrien Halnaut - Visual analysis tools and methods for understanding, optimizaing and creating deep neural networks l'optimisation et l'élaboration de modèles de réseaux de neurones profonds. with David Auber
  • 2019-2022: Korlan Rysbayeva - Data analysis and visualization for soil depolution reports with Nicholas Journet
  • 2015-2018: Rémi Delassus - Machine learning for anomaly detection in open data: application to cartography with Guy Mélançon

Invited PhD Student

  • 2015 (15 days) : Paulo Pisani (Univ Sao Paulo) - Enhanced Template Update: Application to Keystroke Dynamics
  • 2015-2016 (7 months) : Lamis Gouhalmi (Univ. Constantine 2) - Metaheuristics for multibiometrics

Interns

  • 2022: Alexey Zhukov -- Evaluation de méthodes d’explications
  • 2021: Zahra CARN -- IADIFF: Compute and Visualise Differences between Decision Trees
  • 2020: Gaudéric Gumbs -- Visualization for tweets attribution algorithms
  • 2019: Quoc Truong Le - Deep reinforcment learning for graph representation analysis
  • 2018: Gaïzka Alçuyet -- Visualisation de réseaux de neurones profond - Stage M2 ou ingénieur
  • 2018: Lamiaa Snoussi -- Utilisation de CNN pour la prédiction d’usage de systèmes de vélos libre service
  • 2014: Rodrigo Ruiz Díaz -- Méthodes et algorithmes pour la visualisation des performances de systèmes d’authentification biométrique

Current Research

Visualization for deep learning

Large Graph Visualisation

Graph are mathematical objects used in various fields in order to represent relations between objects. However when these graphs contain several thousands or million of nodes and edges their representation with the node-link diagram is not efficient (there are lots of overlaps or computational complexity forbid to draw such graph in acceptable timing). We aim at developing algorithms which compute the layout of a large graph in node-link diagram representation which are computationally efficient and which reduces visual clutter.

Keystroke dynamics

The aim of keystroke dynamics is to recognize users on the way they type on their keyboard. We are interested in static authentication (during the login phase of the user on its computer). We have provided some of the biggest publics datasets on keystroke dynamics for the community.

Template update

Performance of biometric systems degrades through time. The aim of template update is to take into consideration the evolution of the biometric data in order to keep reasonable performance. We have shown that the performance degradation through time of keystroke dynamics system is quite important whereas it as never been taken into account on the evaluation of existing authentication algorithms. We have proposed a template update algorithm which performs better than the self-update system often used in the literature

Previous Research

Securisation of mobile payments

In the MASSIF project, I participate to the adaptation of the tools developed by the partners in order to match the use case which interests France Telecom (securisation of mobile payment thanks to advanced SIEM). My research theme consists in automatic detection of fraud in mobile payment using the transactions events.

Multibiometrics and soft-biometrics

The aim of multibiometrics is to improve the recognition accuracy of biometric systems by fusing them. We have experimented various fusion scheme through fusion functions configured with genetic algorithm or created with genetic programming. The proposed fusion functions perform better than the classic weighted sum of the state of the art. Thus face recognition and keystroke dynamics can be fused together in order to obtain better performance for the best performing sub system ; it allows to obtain cheap and well accepted multibiometrics systems. We have also shown that it is possible to determine the gender of an individual through its way of typing a fixed text (good performances) and any kind of text (lower performances). It is also possible to automatically categorize individuals thanks to their age.

Voxel classification from HARDI data

The understanding of neurodegenerative diseases undoubtedly passes through the study of human brain white matter fiber tracts. To date, diffusion magnetic resonance imaging (dMRI) is the unique technique to obtain information about the neural architecture of the human brain, thus permitting the study of white matter connections and their integrity. However, a remaining challenge of the dMRI community is to better characterize complex fiber crossing configurations, where diffusion tensor imaging (DTI) is limited but high angular resolution diffusion imaging (HARDI) now brings solutions. We investigate the development of both identification and classification process of the local water diffusion phenomenon based on HARDI data to automatically detect imaging voxels where there are single and crossing fiber bundle populations.


The publication list of these works is available here.