Effect of biased feedback in BCI presented at BCI Meeting ’21 conference

Last week the BCI Meeting was held, the major international conference on brain-computer interfaces. As with many other conferences, this edition was the first to be held virtually (with great results, organizers managed to maintain the interactivity between participants, one of the main reasons for a conference to be held in the first place).

Jelena Mladenović presented there our work on the effect of a biased feedback during a motor imagery BCI task. During the experiment, participants had to control a racing game by imagining right hand or left hand movements. Depending on the recognized brain pattern, the character would go either left or right, catching fish. I must mention at this point that said character is a hungry pingouin, Tux from Extreme Tux Racer.

Glimpse at a BCI Tux racing game (with an over-enthusiast participant)

There was three conditions during the experiment: the position of the character could be either positively biased (it was easier to reach targets), negatively biased (harder to to do), or there was no bias (control condition). The main outcome of the experiment is that depending on users’ profiles (e.g. if they are prone to anxiety), the bias could be either helpful or detrimental in terms of performance, learning or flow (an optimal state we can get into while fulfilling a task).

We are in the process of publishing the full paper, if you want to know more about the study the preprint (the version before peers review a paper) is available at: https://hal.inria.fr/hal-03233170.

You can also watch her 10 minutes talk summarizing the study right here:

Jelena’s talk during the vBCI 2021 conference

With this talk Jelena was awarded with a Best Presentation award − in the “non-invasive” category, because in this conference these is also cutting-edge research on invasive techniques, see all the abstracts.

This work was done in collaboration with Jérémie Mattout from Inserm (CRNL COPHY team) and Fabien Lotte from Inria (Potioc team), and we hope to continue our investigations in the foreseeable future (this is already our second study on the topic, previous publication here).

Teaser: we also used the data gathered during this experiment to investigate if it would be possible to automatically select the best bias over the course of a BCI application. And it looks like it, even with a simple selection algorithm. Check Jelena’s thesis for first insights (while the whole piece deserved to be read, this particular subject rests in chapter 5).