Back

Decoding accuracies as well as ERP amplitudes do not show between-task correlations

Bonasch, H.; Ehinger, B. V.

2023-05-23 neuroscience
10.1101/2023.05.21.541632 bioRxiv
Show abstract

Abstract1Some participants consistently show large, and others small activations in Electroencephalography (EEG) and other neuroimaging studies. Similarly, decoding accuracies in Brain-Computer-Interface (BCI) vary between subjects, in extreme cases labelled "BCI-Illiteracy". Here, we investigate whether a switch of task within an event-related design could be sufficient to alleviate low performance. We compare event-related-potentials (ERP) component amplitudes, as well as offline balanced decoding-accuracy based on deep convolutional networks, between seven event-related tasks. ERP effect amplitudes and decoding accuracies were correlated within all tasks, but not between any pairwise tasks. Further, 39/40 subjects had above average performance in at least one task. Two cautious conclusions can be drawn, with the appropriate limitations of power (n=40) and the caveats of interpreting null-findings: 1) The lack of effect amplitude correlations shows that between-subject variability cannot be purely explained by a task-agnostic effects like skull thickness. 2) The lack of decoding accuracy correlations shows promise for ERP-based BCIs: replacing the task could be an effective way to combat "BCI-Illiteracy".

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.