Fig. 5: Behavioural factors decoded from cortical information.

The design of each task was clustered according to psychometric, motor and perceptual dimensions. a Psychometric class defined based on the behavioural factor structure. b Motor class. Dynamic interaction tasks require multiple perceptual-mental-action cycles for completion, e.g. the self-ordered search (SO) task. Sequence tasks require the perceptual encoding of a spatial or temporal stimuli sequence coupled with a motor sequence that reflects that encoding. Finally, static tasks require encoding a static stimuli that requires some mental operation culminating in a forced choice response. c Perceptual-visual class. Composed of spatial configuration, digit, object and word stimuli. A video of each task is provided in the Supplementary movies 1–12. d–f Task to class association showing how each set of classes has a distinct mapping across the tasks. g–i Connectivity/dFC (top row) and voxel-wise activation/BA (bottom row) multi-class classification distribution accuracies (F1-score). Grey colour representing null model, which is an approximation of chance based on shuffled response vectors. Dark colour representing true model performance on the held-out test sample and light colour the random permutation of proportional task to class assignments. p values indicate the significance of the distance of the mean true distribution from the permuted distribution using empirical p value approximation. Overall performances are substantially better than randomly clustered tasks. Behavioural-psychometric and motor classes performed better than visual classes. j–l Scatter plot of per class f1-minor accuracies. Y-axis represent activation models and X-axis connectivity models, showing that internally the models are not affected by the class imbalance. Source data are provided as a Source Data file.