Extended Data Fig. 6: Off states were not explained by an alternate population code, rhythmic activity, or background noise.
From: Intermittent rate coding and cue-specific ensembles support working memory
(A) Cross-validated classifier confidence, binned by training and test state. Left: classifier confidence when trained on Off states and tested on Off and On states. Right: classifier confidence when trained on On states and tested on Off states. Violin plots show bootstrap across sessions. (B) Mean phase-state relationship (z-scored) as a function of frequency. Gray traces show individual sessions (N = 25). Values reflect absolute angular difference between the mean phase of On and Off states, normalized by the null (randomization test). (C) Standard deviation of the background noise, in microvolts, averaged across all On and Off states. Small circles: individual session means; large circles: grand means. Background noise did not differ across On and Off states (p = 0.339, sign-rank).
