[Imageworld] post-doc: Classification of EEG signals by searching for recurrent patterns and learning methods for applications in BCI with a reduced number of electrodes

[Isabelle Bloch]

Brain-computer interfaces (BCI) are investigated since more than 10 years for medical applications. Recently, several projects have been launched for large public applications of BCI, such as mobile applications or video games, but there are still open questions to be solved to allow a simple usage. These questions include the human diversity (morphology and neural activity), the reduction of the number of electrodes, and the localization of sources.
The objective of this study, which contributes to the WHIST-Lab research activities, will address the classification of acquired brain signals, using non supervised learning, so as to make the method adaptive for each subject. This will involve the search for recurrent patterns, which are likely to correspond to the same activity. Simple similarity measures could be first used, and then methods based on sparse representations such as LARS (Mairal, 2010) will be developed. Finally, methods for determining the minimal number of electrodes and their position will be investigated. The aim is to find the right balance between the easiness of use and the quality of classification performances.

More information on http://perso.telecom-paristech.fr/~bloch/postDocBCI.html