Fig. 8: Temporal dynamics and stability.

Information (expressed as time-resolved percent explained variance; solid lines) about factors target duration, stimulus protocol, and their interaction carried by the neuronal population across time (A). Dotted lines show the percent explained variance for shuffled trial labels, shaded areas the SEM over resamples. B Time-resolved SVM classifier trained and tested with 200 ms bins of neuronal activity for the entire population, starting from pre-cue onset until 1200 ms after wait onset. Straight dashed lines mark the onset of the instruction stimulus period and the start and end of the motor planning period. The area outlined by the thick contour line corresponds to the temporal cluster of time bins significantly above chance level (~ 33%; cluster permutation test, see Methods for details). Dashed contour lines indicate different levels of accuracy (35–65% in steps of 10%). C Time-resolved SVM classifier trained and tested using neuronal activity from the wait period of “time” neurons. Activity during the wait period was split into 20 time equally-spaced time-bins (i.e., 20 bins of 60 ms activity for 1500 ms trials, 120 ms bins for 3000 ms trials, and 240 ms bins for 6000 ms trials). For each bin, an SVM model was trained on to distinguish the target duration, then used to predict the target duration of new activity in every other time bin.