RT Journal Article
SR Electronic
T1 Dynamics of brain activity reveal a unitary recognition memory signal
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 165225
DO 10.1101/165225
A1 Christoph T. Weidemann
A1 Michael J. Kahana
YR 2017
UL http://biorxiv.org/content/early/2017/07/19/165225.abstract
AB A repeated encounter with a person or object frequently elicits feelings of familiarity and recollections of previous interactions. Dual-process theories of recognition memory assume that these independently contribute to recognition decisions. We ask whether the memory system integrates available evidence into a unitary evidence signal, or if, instead, different types of mnemonic evidence separately drive distinct routes to recognition. To address this question, we quantified neural evidence for recognition decisions as a function of time using multivariate classifiers trained on spectral EEG features. Assuming distinct temporal profiles for different types of mnemonic evidence, a classifier trained on neural features from multiple time bins during the recognition period should be able to capitalize on independent signals from these evidence sources, if these are indeed kept separate. Instead, we found practically identical classifier performance for classifiers trained on a small portion of the recognition period compared with those also trained on neural features from previous time bins. Recollection has been frequently linked to recall suggesting that a comparison between previously recalled and not previously recalled targets should highlight differences between familiarity- and recollection-based recognition decisions assuming such a distinction is meaningful. Separating classifier performance for targets that were previously recalled and those that were not revealed qualitatively similar increases of classifier performance with time, consistent with different strengths (rather than different kinds) of evidence driving the recognition decisions. These results, along with a strong correspondence between classifier outputs and task performance, firmly link recognition decisions to other types of decisions under uncertainty, which are commonly assumed to rely on a unitary evidence signal differentiating between the response options.Author summary Dual-process models of recognition memory propose that two independent sources of evidence (familiarity and recollection) drive recognition decisions. Usually these are assumed to give rise to different kinds of recognition decisions, with individual “old” responses reflecting either a familiarity or a recollection signal. Alternative accounts posit that all available evidence is integrated into a unitary “memory strength” signal. Using multivariate (“machine learning”) classifiers, we quantified the neural evidence for recognition decisions across various partitions of the recognition period to distinguish between these single- and dual-route accounts of recognition memory. The performance of classifiers that are trained on neural features from a small portion of the recognition period should only match the performance of similar classifiers that also incorporate information from previous time periods to the extent that neural evidence indexes a unitary evidence signal integrating over all available evidence. Compatible with this prediction, we found performance for these two types of classifiers to be practically identical, supporting the view that each recognition decision makes use of all available evidence.