A neural-level model of spatial memory and imagery
- University College London, United Kingdom
Figures
Figure 1
Simplified model schematic.
(A) Processed sensory inputs reach parietal areas and support an egocentric representation of the local environment (in a head-centered frame of reference). Retrosplenial cortex uses current head or …
Figure 2 with 1 supplement
Receptive field topology and visualization of neural activity.
(A1) Illustration of the distribution of receptive field centers (RFs) of place cells (PCs), which tile the environment. (A2) Receptive fields of boundary responsive neurons, be they allocentric …
Figure 2—figure supplement 1
Caption: Illustration of single cell coding in the retrosplenial transformation circuit.
Shaded areas indicate the Parietal Window (PWb), the transformation circuit, head direction modulation, and boundary vector cells (BVCs). Example cells are represented as stylized firing rate maps …
Figure 3
The agent model and population snapshots for object representations.
(A) Top panel: The egocentric field of view of the agent (black arrow head). Purple boundaries fall into the forward-facing 180 degree field of view and provide bottom-up drive to the parietal …
Figure 4
The BB-model.
‘Bottom-up’ mode of operation: Egocentric representations of extended boundaries (PWb) and discrete objects (PWo) are instantiated in the parietal window (PWb/o) based on inputs from the agent model …
Figure 5
(A) Bottom-up mode of operation. Population snapshots at the moment of encoding during an encounter with a single object in a familiar spatial context. Left to right: PWb/o populations driven by …
Figure 6
Firing fields of object vector cells.
(A) Firing rate maps for representative object vector cells (OVCs), firing for objects with a fixed allocentric location and direction relative to the agent. Object locations superimposed as green …
Figure 7
Papez’ circuit lesions.
(A) In the bottom-up mode of operation (perception), a lesion to the head direction circuit removes drive to the transformation circuit and consequently to the boundary vector cells (BVCs) and …
Figure 8
Correlation of neural population vectors between recall/imagery and encoding.
(A) In the intact model, OVCs and place cells exhibit correlation values close to one, indicating faithful reproduction of patterns. (B) Random neuron loss (20% of cells in all populations except …
Figure 9
Detection of moved objects via OVC firing mismatch.
(A) Two objects are encoded from a given location (left). After encoding, object one is moved further North. When the agent returns to the encoding location, the perceived position of object one …
Figure 10
‘Top-down’ activity and ‘trace’ responses.
(A) An environment containing a small barrier (red outline) has been encoded in the connection weights in the MTL, but the barrier has been removed before the agent explores the environment again. (B…
Figure 11
Inspecting scene elements in imagery.
The agent encounters two objects. (A) Activity in PWo (left) and OVCs (right) populations when the agent is attending to one of the two objects during encoding. Both objects are encoded sequentially …
Figure 12
Mental navigation with grid cells.
Left to right: allocentric agent position (black triangle) and recent trajectory (black dashed line); PWo, OVC, and PC population snapshots; GC input to PCs (i.e. GC firing rates multiplied by …
Figure 13
Planning, taking and imaging a trajectory across an unexplored area.
The agent is located in an environment where the direct trajectory between two salient locations (purple dots, left column) covers an unexplored part of the environment. PCs potentially firing in …
Author response image 1
Example of pattern comparison via correlations of population vectors from simulation 1.0 (object cued recall).
White bars show the correlation between the neural patterns during imagery/recall and those during encoding (RvE), while black bars show the average correlation between the neural patterns during …
Videos
Video 1
Surface plots (heat maps) visualize theneural activity of populations of cells.
The video shows a visualization of the simulated neural activity in the retrosplenial transformation circuit as a simulated agent moves in a simple, familiar environment (See Figure 2-figure …
Video 2
This video shows a visualization of the simulated neural activity as the agent moves in a familiar environment and encounters a novel object.
The agent approaches the object and encodes it into long-term memory. Upon navigating past the object the agent initiates recall, reinstating patterns of neural activity similar to the patterns …
Video 3
This video shows the same scenario as Video 2 (object-cued recall), however, with 20% randomly chosen lesioned cells per area.
The agent moves in a familiar environment and encounters a novel object. The agent approaches the object and encodes it into long-term memory. Upon navigating past the object, the agent initiates …
Video 4
This video shows the same scenario as Video 2 (object-cued recall), however, with firing rate noise applied to all neurons (max. 20% of peak rate).
The agent moves in a familiar environment and encounters a novel object. The agent approaches the object and encodes it into long-term memory. Upon navigating past the object the agent initiates …
Video 5
This video shows a visualization of the simulated neural activity as the agent encounters an object and subsequently tries to engage recall similar to Simulation 1.0 (Video 2).
However, a lesion to the head direction system (head direction cells are found along Papez' circuit) precludes the agent from laying down new memories, because the transformation circuit cannot …
Video 6
This video shows a visualization of the simulated neural activity as the agent moves through an empty environment and tries to engage recall of a previously present object.
A lesion to the head direction system (head direction cells are found along Papez' circuit) has been implemented similar to Simulation 1.1 (Video 5). The agent is supplied with the connection …
Video 7
This video shows a visualization of the simulated neural activity in a reproduction of the object novelty paradigm of Mumby et al., 2002; detecting that one of two objects has been moved).
The agent is faced with two objects and encodes them (sequentially) into memory. Following some behavior one of the two objects is moved. Note, in real experiments the animal is removed for this …
Video 8
should be compared to Video 7.
It shows a reproduction of the object novelty paradigm of Mumby et al., 2002; detecting that one of two objects has been moved). The agent is faced with two objects and encodes an association …
Video 9
This video shows a visualization of the simulated neural activity as the agent moves in a familiar environment.
However, a previously present boundary has been removed. The agent is supplied with a periodic (akin to rodent theta) modulation of the top-down connection weights (please see main text). The …
Video 10
This video shows a visualization of the simulated neural activity as the agent moves in a familiar environment.
However, a previously present (and encoded) object has been removed. The agent is supplied with a periodic (akin to rodent theta) modulation of the top-down connection weights (please see main …
Video 11
This video shows a visualization of the simulated neural activity as the agent sequentially encodes two objects into long-term memory.
Upon navigating past the objects the agent initiates recall, cueing with the first object. The OVC representations of both objects are bound to the same place cells. These place cells thus generate …
Video 12
This video shows a visualization of the simulated neural activity as the agent performs a complex trajectory and encodes three objects into long-term memory along the way.
Upon navigating past the third object the agent initiates recall, cueing with the first object, and subsequently performs mental navigation (imagined movement in visuo-spatial imagery) with the help …
Video 13
This video shows a visualization of the simulated neural activity as the agent performs mental navigation across a blocked shortcut.
Newly recruited cells in the hippocampus exhibit activity reminiscent of preplay. Upon removal of the barrier the agent traverses the shortcut and associates the newly recruited hippocampal cells …
Tables
Table 1
List of simulations, their content, corresponding Figures and videos
https://doi.org/10.7554/eLife.33752.009| Simulation no. | Content | Related figures | Video no. |
|---|---|---|---|
| 0 | Activity in the transformation circuit | Figure 2—figure supplement 1 | 1 |
| 1.0 | Object-cued recall | Figures 5 and 6,8A | 2 |
| 1.0n1 | Object-cued recall with neuron loss | Figure 8B | 3 |
| 1.0n2 | Object-cued recall with firing rate noise | Figure 8C | 4 |
| 1.1 | Papez’ circuit Lesion (anterograde amnesia) | Figure 7A | 5 |
| 1.2 | Papez’ circuit Lesion (retrograde amnesia) | Figure 7B | 6 |
| 1.3 | Object novelty (intact hippocampus) | Figure 9A | 7 |
| 1.4 | Object novelty (lesioned hippocampus) | Figure 9B | 8 |
| 2.1 | Boundary trace responses | Figure 10A,B,C | 9 |
| 2.2 | Object trace responses | Figure 10D | 10 |
| 3.0 | Inspection of scene elements in imagery | Figure 11 | 11 |
| 4.0 | Mental Navigation | Figure 12 | 12 |
| 5.0 | Planning and short-cutting | Figure 13 | 13 |
Appendix 1—table 1
Model Parameters.
Top to bottom: α, β sigmoid parameters; φ connection gains; Φ constants subtracted from given weight matrices (e.g. PC to PC connections) to yield global inhibition; bath parameters; range …
| α | 5 |
|---|---|
| β | 0.1 |
| αIP | 50 |
| βIP | 0.1 |
| φPWb-TR | 50 |
| φTR-PWb | 35 |
| φTR-BVC | 30 |
| φBVC-TR | 45 |
| φHD-HD | 15 |
| φHD-IP | 10 |
| φHD-TR | 15 |
| φHDrot | 2 |
| φIP-TR | 90 |
| φPC-PC | 25 |
| φPC-BVC | 1100 |
| φPC-PRb | 6000 |
| φBVC-PC | 440 |
| φBVC-PRb | 75 |
| φPRb-PC | 25 |
| φPRb-BVC | 1 |
| φGC-PC | 3 |
| φPWo-TR | 60 |
| φTR-PWo | 30 |
| φTR-OVC | 60 |
| φOVC-TR | 30 |
| φPC-OVC | 1.7 |
| φPRo-OVC | 6 |
| φPC-PRo | 1 |
| φOVC-PC | 5 |
| φOVC-oPR | 5 |
| φPRo-PC | 100 |
| φPRo-PRo | 115 |
| φinh-PC | 0.4 |
| φinh-BVC | 0.2 |
| φinh-PRb | 9 |
| φinh-PRo | 1 |
| φinh-HD | 0.4 |
| φinh-TR | 0.075 |
| φinh-TRo | 0.1 |
| φinh-PW | 0.1 |
| φinh-OVC | 0.5 |
| φinh-PWo | 1 |
| ΦPC-PC | 0.4 |
| ΦBVC-BVC | 0.2 |
| ΦPR-PR | 9 |
| ΦHD-HD | 0.4 |
| ΦOVC-OVC | 0.5 |
| ΦPRo-PRo | 01 |
| PWbath | 0.1 |
| PWbath | 0.2 |
| TRbath | 0.088 |
| Object enc. threshold | 18 cm |
| Object enc. Threshold (3.1) | 36 cm |
| lGC-resPC | 0.65*10^−5 |
| lresPC-BVC | 0.65*10^−5 |
| lBVC-resPC | 0.65*10^−5 |
| SGC-resPC | 3% |
| SresPC-resPC | 6% |
| σρ | (r + 8) * σ0 |
| σ0 | 0.08 |
| σϑ | 0.2236 |
| NPC | 44 × 44 |
| NBVC | 16 × 51 |
| NTRb/o | 20 × 16×51 |
| NOVC | 16 × 51 |
| NPRb/o | Dependent on simulation environment |
| NPWb/o | 16 × 51 |
| NIP | 1 |
| NHD | 100 |
| NGC | 100 per module |
| Nreservoir | 437 |
Additional files
-
Transparent reporting form
- https://doi.org/10.7554/eLife.33752.030
