Figure S6.

Sequenceness Analysis with Classifiers Trained at Different Times, Related to Figures 2 and 4

a, Sequenceness using classifiers trained at different times relative to stimulus onset (110 ms – 300 ms). This is from resting data after applied learning in Study 2. b, Scatterplot of the sequenceness at 40 ms time lag as a function of classifier training times in Study 2 during rest period after applied learning. 200 ms is the training time used throughout the current study. c, Scatterplot of the sequenceness at 40 ms time lag as a function of classifier training time in Study 1 during rest period after value learning. d, A similar result was obtained by re-analyzing data from Kurth-Nelson et al. (2016). Notably, there is a preference for replay of a particular time-slice of the representation (at 200 ms) in all 3 studies. That is, although the representation of each stimulus evolves over a period of approximately 500 ms, only one time-slice (at 200 ms) reliably replays during rest (with current data analysis approaches). This time-slice (200 ms) is consistent across 3 independent datasets.