This is an online seminar. Registration is required.
Registered participants will receive details for online access a day before the seminar.
Title: Applications of machine learning in non-invasive brain-computer interfacing and functional neuroimaging
Abstract: Brain-computer interfaces (BCIs) are tools that convert brain activity into artificial outputs to bypass natural neuromuscular outputs and thereby directly change the interaction between the brain and its environment. This talk covers non-invasive BCIs for classifying mental tasks and decoding movement trajectories, and how machine learning can be used to overcome specific challenges in these applications.
The first application focuses on inter-subject transfer and co-adaptive learning. In an EEG study with naïve BCI users, we demonstrated that a pre-trained classification restricted Boltzmann machine could reliably detect the users’ distinct imaginations and that the two-learner system (model and user) could significantly improve the task performance through feedback training and mutual adaptation.
In the second application, we propose a proof-of-concept, non-invasive BCI to infer movement trajectories of actions in real-time. To do so, it was necessary to develop neuroimaging tools which are sensitive to movement trajectory related effects in the EEG and MEG activity and invariant to co-varying eye artifacts. The new tools allowed us to corroborate and extend the findings of previous works by identifying and characterizing two cortical networks that encode directional and non-directional kinematic information during discrete and continuous movements. The methodological and conceptual advances informed the design of a proof-of-concept BCI for movement control. In two feasibility studies, we found that the proposed BCI decoded movement trajectories in real-time with moderate congruence.
|Date||July 15, 2020 (Wed) 15:00 - 16:00|