Shared Representational Formats for Information Maintained in Working Memory and Information Retrieved from Long-Term Memory - PubMed (original) (raw)
Shared Representational Formats for Information Maintained in Working Memory and Information Retrieved from Long-Term Memory
Vy A Vo et al. Cereb Cortex. 2022.
Abstract
Current theories propose that the short-term retention of information in working memory (WM) and the recall of information from long-term memory (LTM) are supported by overlapping neural mechanisms in occipital and parietal cortex. However, the extent of the shared representations between WM and LTM is unclear. We designed a spatial memory task that allowed us to directly compare the representations of remembered spatial information in WM and LTM with carefully matched behavioral response precision between tasks. Using multivariate pattern analyses on functional magnetic resonance imaging data, we show that visual memories were represented in a sensory-like code in both memory tasks across retinotopic regions in occipital and parietal cortex. Regions in lateral parietal cortex also encoded remembered locations in both tasks, but in a format that differed from sensory-evoked activity. These results suggest a striking correspondence in the format of representations maintained in WM and retrieved from LTM across occipital and parietal cortex. On the other hand, we also show that activity patterns in nearly all parietal regions, but not occipital regions, contained information that could discriminate between WM and LTM trials. Our data provide new evidence for theories of memory systems and the representation of mnemonic content.
Keywords: fMRI; long-term memory; multivariate pattern analysis; working memory.
© The Author(s) 2021. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permission@oup.com.
Figures
Figure 1
Memory tasks and perceptual task presented to participants in the scanner. (A) The two memory tasks were identical except for the memory cue—in the WM task, participants saw a dot indicating the position along an annulus, along with an irrelevant clip art item; in the LTM task, participants instead saw a previously studied clip art item that they had learned to associate with a spatial position during behavioral training. After an 11.5-s delay period, participants reported the remembered spatial position by rotating a randomly placed dot to the correct position. (B) An independent task was used to map the position selectivity of voxels in retinotopically defined regions of visual and parietal cortex. Participants detected an occasional dimming of a checkerboard stimulus that appeared at randomly ordered locations along a ring. This independent task was used to train the IEM, which was then tested on the memory tasks. (C) Example spatial representations and their corresponding fidelity values, a single number which characterizes the quality of the representation of the remembered position (always set to 0° here). Each spatial representation was plotted in polar space. The fidelity metric is equivalent to the length of the bold horizontal vectors. Dark purple: The best model-based spatial representation is narrow and centered exactly at 0°. Light purple: A broader spatial representation has a shorter mean vector, capturing the fact that less “energy” is at 0°. Dark green: A spatial representation that is slightly offset from zero has a short _x_-component and therefore a lower fidelity value, as compared to a representation centered at zero. Light green: An inverted spatial representation has a mean vector that points in the opposite direction, resulting in a negative fidelity value.
Figure 2
Retinotopic parietal ROIs (outlines) and lateral parietal ROIs (colored) for the left hemisphere of an example participant.
Figure 3
Memory recall across both scanner tasks is similar. (A) The distribution of memory recall errors for an example participant. We fit a mixture model to this distribution for every participant, where the Gaussian distribution characterized the variability of recall (SD) and the uniform distribution characterized the likelihood of recall (P(recall)). (B) The mixture model fit parameters for each subject and task. The mean across participants and 95% CIs are shown in black.
Figure 4
Model-based representations of remembered spatial positions over time for both memory tasks. Error bars are participant-resampled 95% CIs. (A) The timecourse of model-based spatial representations using V1 data, averaged across participants (0 s is stimulus onset). Remembered position was represented similarly for WM and LTM late in the delay period (8–14 s). (B) Representational fidelity timecourses for all retinotopic areas we analyzed. CIs that intersect with 0 are not significant. dLatIPS, dorsolateral IPS; vLatIPS, ventrolateral IPS; AnG, AnG.
Figure 5
Model-based representations for different combinations of training and testing data, taken from the end of the delay period (8–12 s). For each plot, we ran a permuted two-way ANOVA of ROI by memory task (or for the sensory control, a one-way ANOVA across ROIs). (A) Train on the sensory task and test memory, as a test of sensory reinstatement. (B) As a control, train and test within the sensory task. (C) Train and test within each memory task, to examine task-specific representations. (D) Cross-train memory tasks, e.g. train WM and test LTM, to examine memory representations that are general across tasks. In all four cases, there was a significant main effect of ROI (P < 0.001). We found a significant two-way interaction for both the independent training set (P = 0.003) and the cross-training procedure (P = 0.001), but no main effect of task. Training within memory tasks yielded a borderline significant effect of memory task (P = 0.052), as the WM task generally resulted in higher fidelity than the LTM task. Asterisks indicate significant information after FDR correction across all ROIs and both memory tasks, q = 0.05.
Figure 6
Task decoding in the early and late epochs of the delay period in each ROI. Colored dots show data from each individual subject, and error bars are 95% CIs estimated with bootstrapping. Classifier accuracy is significantly above chance (*) if it passes a one-tailed a permutation test, FDR-corrected across ROI and temporal epoch (q = 0.025). Unlike the occipital ROIs, the activation patterns in regions of parietal cortex (V3AB, IPS0, etc.) resulted in above chance decoding accuracy in the late epoch.
References
- Atkinson RC, Shriffin RM. 1968. Human memory: a proposed system and its control processes. In: Spence KW, Spence JT, editors. Psychology of learning and motivation. New York: Academic Press, pp. 89–195.
- Awh E, Jonides J. 2001. Overlapping mechanisms of attention and spatial working memory. Trends Cogn Sci. 5:119–126. -PubMed
- Awh E, Vogel EK, Oh SH. 2006. Interactions between attention and working memory. Neuroscience. 139:201–208. -PubMed
- Baddeley A. 2000. The episodic buffer: A new component of working memory? Trends Cogn Sci. 4:417–423. -PubMed
Publication types
MeSH terms
LinkOut - more resources
Full Text Sources