@vclay leads a deep dive into grid cell literature. Some topics discussed include how brains continually update existing models with new information, create separate models when necessary, and combine old and new knowledge during inference. The team also discussed evidence that visual landmarks can initialize and anchor grid cells after a single exposure. The conversation also focused on grid-cell remapping, the gradual merging of reference frames when barriers are removed, rapid switching between room-based and object-based reference frames, and the idea that grid-cell modules can operate at different spatial scales, allowing them to independently map local features and entire environments.
Main Video
Summary Video
0:00 Introduction
0:24 Questions and Problems We’re Trying to Solve
3:12 2024 Paper: One-shot Entorhinal Maps Enable Flexible Navigation in Novel Environments
4:46 Key Takeaway: Landmarks Influence Grid Cell Properties Within a Single Exposure
9:57 Key Takeaway: Fixed Landmark-to-Grid-Cell Model Can Predict Grid Cell Activity for a Held-out Landmark Combination
13:09 Grid Cell Modules at Different Scales
17:13 Landmark Changing can Distort Grid Cell Space But Downstream Fast Plasticity Allows for Quick Behavior Adaptation
23:10 Types of Remapping: Rate-Changes (Color), Translation (Color or Ordor), Translation + Orientation (Strong Metric Changes to the Environment)
28:40 Paper: Integration of Grid Maps in Merged Environments
35:07 Paper: Grid Cells Form a Global Representation of Connected Environments
38:54 What Does This Mean for Monty?
42:52 There Can Be Multiple Reference Frames For the Same Environment, Depending on the Behavioral State or Task Due to Spontaneous Remapping
50:34 Monty Will Have Learning Modules at Different Scales
54:25 Last Paper: Grid Cells Accurately Track Movement During Path Integration-Based Navigation Despite Switching Reference Frames
1:06:25 High-Level Takeaways
1:11:57 Final Thoughts