To permit the identification of underperformance localised to more complex subpopulations defined by the interactions of multiple factors, we used an autoencoder to embed participants in a two-dimensional latent representational space that compactly described their high-dimensional similarities and differences ...
(A) Latent-space samples were passed through decoder layers of the autoencoder (each session) to reconstruct high-dimensional activity patterns. Single-session average right thumb γH activity across the grid for the three BCI experiments in B1 are shown after reconstruction. It can be seen that...