Abstract
Working memory (WM) is an online memory system that is critical for holding information in a rapidly accessible state during ongoing cognitive processing. Thus, there is strong value in methods that provide a temporally-resolved index of WM load. While univariate EEG signals have been identified that vary with WM load, recent advances in multivariate analytic approaches suggest that there may be rich sources of information that do not generate reliable univariate signatures. Here, using data from 4 published studies (n = 286 and >250,000 trials), we demonstrate that multivariate analysis of EEG voltage topography provides a sensitive index of the number of items stored in WM that generalizes to novel human observers. Moreover, multivariate load detection can provide robust information at the single-trial level, exceeding the sensitivity of extant univariate approaches. We show that this method tracks WM load in a manner that is (1) independent of the spatial position of the memoranda, (2) precise enough to differentiate item-by-item increments in the number of stored items, (3) generalizable across distinct tasks and stimulus displays and (4) correlated with individual differences in WM behavior. Thus, this approach provides a powerful complement to univariate analytic approaches, enabling temporally-resolved tracking of online memory storage in humans.
Significance Statement Working memory (WM) is a workspace used to temporarily hold information in mind, and it is critical for complex cognition. Because behavioral measures are influenced by myriad task factors (e.g., response bias), neural measures are critical for characterizing WM maintenance per se and for tracking its involvement in other processes. Here, we used a large dataset to develop multivariate load detection: a new method for tracking active WM storage using the human EEG signal. We show that multivariate load detection is incredibly sensitive and generalizable, predicting load on single trials and generalizing across tasks and observers. Thus, multivariate load detection offers key advances over existing univariate measures and will be useful for both basic and applied research (e.g., brain-computer interfaces).
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Funding: Research was supported by National Institute of Mental Health grant 5R01-MH087214 (to E.A. & E.V.) and Office of Naval Research grant N00014-12-1-0972 (to E.V.). K.A. was supported by National Institute of Mental Health grant 5T32-MH020002.
Conflicts of interest: None.