When we looked more closely at why this type of model did so well, we found it is because it predicts how the CA1 spatial code moves when you change geometry of the space.

We then transformed these predictions into our big similarity matrix, and directly compared model predictions to the representation we observed in CA1. This revealed the winner(s) are based on a specific set of boundary vector inputs to explain remapping.

Next we asked how we could generate predictions of handful of distinct models of cognitive mapping in this task. Thank you very much RatInABox!! @tomnotgeorge.bsky.social

This showed that remapping across animals were very reliable in this task, and becomes more similar across animals with repeated experience (repeating sequences of geometries). It turns out you need to record at least a couple hundred cells in the hippocampus to see these levels of similarity.

When we did this, we saw maps changing in some cool ways. We then asked – how reliable is this pattern of remapping across animals? To address this, we took a representational similarity approach to compare the pattern of remapping across animals.

We know that the hippocampus creates maps of spaces and tasks, and that when you change spaces and tasks the maps change (remapping). One driver of this in the rodent hippocampus is geometry. So we came up with a way to create diverse geometries in a single environment, and record lots of neurons!

One more time!! Presenting my final postdoc project with
@markbrandonlab
at U29 from 8–12 this morning! Come say hi!! #SfN24