Predicting Memory from EEG


Outcome:
TDLC's Eunho Noh and Virginia de Sa at the University of California, San Diego and Grit Herzmann* and Tim Curran at the University of Colorado, Boulder have found that they can predict (with accuracy of 57.2%) whether someone will remember an upcoming picture from the voltage recorded at the scalp (electroencephalography (EEG)) prior to the picture presentation. The prediction can be improved to 59.6% if EEG during picture presentation is also used.

Impact/benefits:
This result could be used to develop improved study systems that monitor the user's brain state and present items to be memorized during predicted "good" brain states. Items that were not deemed well-encoded according to EEG measurements during presentation of the item, could be presented again. Finally extensive use of this system may lead to users becoming better able to get into attentive
brain states that are good for memorizing information. This may be the best benefit of all.

Background/Explanation:
The system looks at both the temporal waveform of the EEG signal during encoding and also the power in different frequency bands prior to item presentation and during encoding. Further analysis shows that the different frequency bands in the signal before the item is presented differentially predict whether the context will be remembered with the item. The higher frequency content (from 25-35Hz) of the EEG signal before item presentation distinguishes between recollection (remembering the item and the context) and familiarity (remembering the item without the context). Similarly the temporal signal in the later period (1-1.4 seconds after item presentation) distinguishes between recollection and familiarity while an earlier period (.4-.8 seconds after presentation) does not.



Noh, E., Herzmann, G., Curran, T. & de Sa, V.R. (2014). Using Single-trial EEG to Predict and Analyze Subsequent Memory. Neuroimage, 84(1):712-723.

* Grit Herzmann is now an Assistant Professor at The College of Wooster.