Brain-Computer Interfaces (BCI) are neurotechnologies that use brain activity (with electroencephalogram-type acquisition machines) to control systems. The individual performs tasks without using muscles. This technology is very promising for many applications (prosthesis control, post-stroke rehabilitation...), but they are still insufficiently reliable. Making them reliable and usable outside the laboratory requires not only improvements on the machine side (e.g.: brain signal analysis algorithms), but also on the user side. Indeed, controlling a BCI is a skill that can be learned and that requires training. Unfortunately, the scientific community still has a poor understanding of how to train users effectively and how to help them acquire the necessary skills. In order to optimize BCI training, it is first necessary to understand and model it, and then to deduce the factors that can be played on in order to improve the user's performance and learning. The objective of my research is to study and statistically model how users learn to produce BCI commands. To do this, I create interpretable computational models that can predict the average performance and progression of BCI users. These models are created using factors from the users themselves such as their personality traits, or characteristics of their EEG activity. They also take into account the characteristics related to the machine learning to decode the EEG signals of the users to transform them into commands, but also feedback. By combining these different models, we hope to be able to predict the performance of BCI users at any time and thus identify the factors that influence it in order to optimize the training of users.

Virtual room URL:
https://u-bordeaux-fr.zoom.us/j/82577055914?pwd=OFVTSGc1dkUya0MvQnlydVZKNmV0dz09