How sturdy is your memory palace? Reliable room representations predict subsequent reinstatement of placed objects

Rolando Masís-Obando, Kenneth A. Norman, Christopher BaldassanobioRxiv 2024 · DOI

TL;DR

This study reveals that the neural "sturdiness" or reliability of a spatial environment, measured before any new memories are formed within it, predicts how well objects placed in that environment will be recalled later. This suggests that stable and distinct neural representations of locations act as better scaffolds for binding and retrieving new information.

What Problem Does This Paper Address?

How do spatial contexts, like rooms in a building, help us remember things? While we know that location is important for memory (think of the "memory palace" technique), it's unclear what makes some locations better memory aids than others. Specifically, can we predict a location's usefulness as a memory scaffold before we even try to store information there? This paper investigates whether the neural stability and distinctiveness of a spatial representation can be quantified and used to predict subsequent memory performance for objects placed within that space.

Methods

The researchers developed a novel experimental paradigm using a custom-built 23-room virtual reality (VR) "memory palace." * Neural Reliability Measurement: Participants first explored the VR environment. Then, during fMRI scanning, they watched videos of each room twice. A "room reliability" score was computed for each room by measuring how similar the neural activity patterns were to themselves across these viewings (stability) versus how similar they were to other rooms (distinctiveness). This was done before any objects were introduced. * Object Encoding and Recall: Participants then returned to VR to learn the locations of 23 new objects, one in each room. Later, while in the fMRI scanner, they recalled these objects. * Object Reinstatement Measurement: A "Retrieved Object Classifier Network" (ROCN) was trained using fMRI data from other participants to identify brain regions that represent specific objects. The activity within this network during recall was used as a measure of "object reinstatement" – how strongly the neural representation of an object was reactivated. * Relating Reliability to Reinstatement: The core analysis involved correlating the pre-learned room reliability scores with the ROCN object reinstatement scores during recall. * Controlling for Room Reinstatement: To ensure the findings weren't just due to participants simply recalling the rooms better, a partial correlation analysis was performed, controlling for the neural evidence of room reinstatement during recall.

Datasets

The study utilized data collected from 30 participants (25 after exclusions) who interacted with a custom-built 23-room virtual reality environment. No external datasets were used for training or evaluation.

Key Insights and Ideas

  • "Sturdy" Spaces are Better Memory Scaffolds: The core idea is that the neural "sturdiness" – the stability and distinctiveness of a location's neural representation – is a key factor in its effectiveness as a memory scaffold.
  • Predicting Memory Potential: It's possible to "audit" a spatial environment's memory-supporting potential by measuring the reliability of its neural representation before encoding new memories.
  • Beyond Group Averages: The effect of room reliability on object recall wasn't just a group phenomenon; individual participants' unique neural representations of rooms also predicted their object recall performance.
  • Binding, Not Just Retrieval: The findings suggest that reliable room representations facilitate better binding of objects to locations during encoding, rather than simply making it easier to retrieve the room itself at test.

Key Results

  • Pre-learning room reliability significantly predicted object reinstatement during verbal recall across widespread brain regions, including parietal cortex, prefrontal cortex, and precuneus.
  • This relationship held true for both group-level reliability (rooms that were reliably represented across participants) and idiosyncratic reliability (rooms that were reliably represented for a specific participant).
  • Crucially, the relationship between room reliability and object reinstatement remained significant even after statistically controlling for room reinstatement during recall. This suggests that the effect is not simply due to better retrieval of reliable rooms, but rather an earlier influence on memory formation.

Limitations and Future Work

  • Negative Relationships: The study observed negative relationships between room reliability and object reinstatement in some lower-level sensory cortices. The authors suggest this might be due to these regions coding for isolated features rather than the overall spatial context, potentially hindering its use as a contextual anchor. Further research is needed to clarify this.
  • Encoding Phase: While the study infers facilitated binding at encoding, direct measurement of binding during the encoding phase is missing. Future work could involve fMRI during the object-location encoding task.
  • Contextual Factors: The authors note that other factors like the number of items per room, temporal variability within a room, or novelty might influence how room reliability scaffolds memory. Future studies could explore these interactions.
  • Method of Loci: The study's paradigm draws inspiration from the method of loci. Future work could investigate whether enhanced room reliability also benefits memory when participants explicitly use the method of loci.

Summary updated

Abstract

Abstract What are the neural properties that make spatial contexts effective scaffolds for storing and accessing memories? We hypothesized that spatial locations with stable and distinctive (i.e., reliable) neural representations would best support memory for new experiences. To test this, participants learned the layout of a custom-built 23-room virtual reality (VR) “memory palace” that they explored using a head-mounted display. The next day, participants underwent whole-brain fMRI while watching videos of the rooms, allowing us to measure the reliability of the neural activity pattern associated with each room. Participants then returned to VR to encode 23 objects placed in each of the 23 rooms and later recalled the rooms and objects during fMRI. We found that our room reliability measure (computed prior to encoding) predicted object reinstatement during recall across cortex; this was driven not only by group-level reliability across participants, but also idiosyncratic reliability within participants. Moreover, this effect did not arise through enhanced retrieval of reliable rooms during recall, since the relationship between reliability and object reinstatement remained significant when controlling for room reinstatement during retrieval; this suggests that, instead, room reliability promotes improved binding of rooms to objects at encoding. Together, these results showcase how the quality of the neural repre-sentation of a spatial context can be quantified and used to ‘audit’ its utility as a memory scaffold for future experiences.
BibTeX
@misc{masisobando2024how,
  author = {Rolando Masís-Obando and Kenneth A. Norman and Christopher Baldassano},
  title = {{How sturdy is your memory palace? Reliable room representations predict subsequent reinstatement of placed objects}},
  year = {2024},
  doi = {10.1101/2024.11.26.625465},
  howpublished = {bioRxiv},
}
Figure