2018-00778 - Ph-D in Computational Modeling of Mental Imagery-based Brain-Computer Interface (BCI) user training
Le descriptif de l’offre ci-dessous est en Anglais

Type de contrat : CDD de la fonction publique

Niveau de diplôme exigé : Bac + 5 ou équivalent

Fonction : Doctorant

A propos du centre ou de la direction fonctionnelle

The Potioc project-team explores new approaches “beyond the mouse” in the field of Human-Computer Interaction. More specifically, we are interested in approaches that favor rich interactions, both regarding interaction possibilities and perceptive feedback. Our objective is to increase immersion and engagement of users with regard to the interaction tasks. Final goals are the stimulation of creativity, improvement of learning or contribution to the well-being of people. For achieving these goals, we focus on the design, development and evaluation of new methods for “popular interaction” targeted at a large variety of users.

Contexte et atouts du poste

Brain computer interfaces (BCI) are communication and control tools that enable their users to interact with computer by using brain activity alone.

A prominent type of BCI is Mental Imagery (MI) BCI, that translates change in brain activity due to mental imagery tasks performed by the user (e.g., imagination of movements or mental calculation) into control commands for a computer.

Using a MI-BCI requires dedicated training, and the more the user practice the better he/she will get at it, i.e. the user's mental commands will be more often correctly recognised by the system. Current BCI are rather unreliable, and one current hypothesis to explain this could be inappropriate user training.

We are thus currently conducting researches to understand this user training to then improve it and make it suitable.

Such researches are conducted as part of ERC starting grant project BrainConquest (https://team.inria.fr/potioc/brainconquest/ – Principal Investigator: Fabien Lotte) at Inria Bordeaux Sud-Ouest, France, in team Potioc (https://team.inria.fr/potioc/).
 

Mission confiée

As part of this research, the goal of this PhD thesis would be to contribute to the understanding of BCI user training by trying to model its various components computationally.

The idea would be to create computational models of BCI user training that could predict the learning rate and the performances of various BCI users, over training time, based on their traits, states and skills over time, and based on the feedback they receive and on the training tasks they perform. Different models will be created to account for different aspects and time scale of learning, e.g., to predict the overall performances, but also the performances variability between sessions/runs, as well as to predict whether a given trial is going to be successful or not. Understanding and modeling how the BCI users exploit and learn from the feedback they receive is also of high interest.

More particularly, we are interested in identifying how different properties of the feedback affect BCI learning and performances. Such properties could include the feedback visual characteristics, update rate, bias, the fact of being positive/negative or positive only, its modality, etc.  By combining these various models (models of performance, of feedback processing, of performance variability across runs and trials, etc.), we hope to be able to successfully predict performances and learning at any time. This would enable us to identify the factors to consider in order to improve user training, and how to manipulate them to optimize such training.

The goal would be to develop these models both for the healthy user population, but also for the BCI end users population, in particular for stroke patients (based on data from such users). Such models could be statistical/probabilistic model. They could be based on regression, hidden markov model, Bayesian networks, control theory or other generative machine learning/statistical tools. They could also be based on computational implementation of theoretical models of BCI performances or models of learning in general, from the psychology literature.

We already have data from multiple BCI users who trained over multiple days with MI-BCI, and for which we measured the traits using psychological questionnaires. We also have data sets from MI-BCI users who trained with different feedback characteristics such as visual or tactile feedback, positive only feedback, and various feedback biases. These data sets will be the basis for creating, testing and validating different initial models.

The PhD thesis could also include designing and running MI-BCI experiments to acquire additional data to refine the initial models (e.g., to include missing factors), to
validate them on different MI-BCI training tasks.

Principales activités

  • Be able to successfully predict performances and learning at any time
  • Develop  models (models of performance, of feedback processing, of performance variability across runs and trials, etc.) both for the healthy user population, but also for the BCI end users population, in particular for stroke patients (based on data from such users).
  • Design and run MI-BCI experiments to acquire additional data to refine the initial models

Compétences

Skills required :

• Modelling, statistical analysis and tools, and/or machine learning

• Python / Matlab programming

• Able to speak, write and work in an English speaking environment

• Skills in neurosciences, psychologie, cognitive science appreciated

• Experience with ElectroEncephaloGraphy (EEG) and/or BCI experiments appreciated

Related literature :

Lotte, F., Larrue, F., & Mühl, C. (2013). Flaws in current human training protocols for spontaneous Brain-Computer Interfaces: lessons learned from instructional design.Frontiers in human neuroscience, 7.
 
Jeunet, C., N’Kaoua, B., Subramanian, S., Hachet, M., & Lotte, F. (2015). Predicting mental imagery-based BCI performance from personality, cognitive profile and neurophysiological patterns. PloS one, 10(12), e0143962.
 
Jeunet, C., N’Kaoua, B, & Lotte, F. (2017). Towards a cognitive model of MI-BCI user training. International Graz BCI conference Lotte, F., & Jeunet, C. (2017). Online classification accuracy is a poor metric to study mental imagery-based BCI user learning: an experimental demonstration and new metrics. In 7th International BCI Conference.
J. Mladenovic, J. Frey, M. Bonnet-Save, J. Mattout, F. Lotte, "The Impact of Flow in an EEGbased Brain Computer Interface", 7th international BCI conference, 2017 Roos, D. F. A. (2017). Modelling BCI Learning.
 
C. Jeunet, F. Lotte, JM. Batail, P. Philip, JA. Micoulaud-Franchi, “Using recent BCI literature to deepen our understanding of clinical neurofeedback: A short review”, Neuroscience, 2018

Avantages sociaux

  • Subsidised catering service
  • Partially-reimbursed public transport

Rémunération

1982€ / month (before taxs) during  the first 2 years, 2085€ / month (before taxs) during the third year.