(A) Latent state dynamics of the gradCPT data collected by Rosenberg et al., 2016 (N = 25). The data consists of three runs, with each run consisting of four 3 min task blocks interleaved with 30 s of fixation. A boxcar plot indicates durations of the task and fixation blocks, shifted in time (5 TR; 5 s) to account for hemodynamic response delay. White areas indicate missing fMRI runs. (B) Fractional occupancies of the four states were calculated during task and fixation blocks separately and averaged across participants. The bar graph indicates the mean of fractional occupancies calculated per participant. The default mode network (DMN), dorsal attention network (DAN), and base states occurred more frequently during task than fixation blocks (paired t-tests, t(24) > 6, FDR-p-values<0.001), whereas the SM state occurred more frequently during fixation blocks (t(24) = 12.565, FDR-p<0.001). We observed a significant main effect of neural states (repeated-measures ANOVA: F(3,72) = 85.479, p<0.001) and an interaction between neural states and block types (F(3,72) = 119.392, p<0.001). The error bars indicate SEM. (C) Latent state dynamics of the Human Connectome Project (HCP) working memory (WM) task runs in left-to-right (LR; N = 110) and right-to-left (RL; N = 108) phase encoding sessions. Each session consists of a single run, with four 2-back WM (25 s), four 0-back WM (25 s), and three fixation blocks (15 s) interleaved, which are indicated in the boxcar plots shifted in time (6 TR; 4.32 s). (D) Fractional occupancies of the four states calculated at moments of 2-back WM, 0-back WM, and fixation blocks. In both sessions, DAN state occurrence decreased (paired t-tests for all block pairs, LR: t(109) > 13, RL: t(107) > 10, FDR-p-values<0.001), whereas the DMN and SM state occurrence increased (LR: t(109) > 3, RL: t(107) > 5, FDR-p-values<0.001), from 2-back to 0-back to fixation blocks as blocks required less cognitive control and memory load. The base state, on the other hand, occurred comparably during the 2-back and 0-back WM task blocks (LR: t(109) = 1.804, FDR-p=0.074, RL: t(107) = 1.486, FDR-p=0.140), whereas it occurred less at fixation blocks (LR: t(109) > 10, RL: t(107) > 9, FDR-p-values <0.001). The repeated-measures ANOVA showed the main effect of the neural states (LR: F(3,327) = 117.445, p<0.001; RL: F(3,321) = 126.702, p<0.001) and the interaction effect (LR: F(6,654) = 261.987, p<0.001; RL: F(6,642) = 149.786, p<0.001).