Fig. 7.

Attention decoding performance on 20 ​s trials: SI-GEVD filtered data vs raw data vs DSS filtered data (trained only on 3 repetitions per subject (DSS-3reps), or trained on common stimulus presentations across subjects (DSS-universal)). The blue dotted line indicates the significance level (95 percentile) for performance above chance level. Each boxplot is built from the mean decoding accuracy per subject, also indicated by colored points. Improvements in mean decoding accuracies of subjects when SI-GEVD filtering are indicated by green lines. In the plot, comparisons between methods were done using Wilcoxon’s signed-rank tests (with Bonferroni correction): ‘***’ for $p<0.001$, ‘****’ for $p<0.0001$.