HTM School 14: Grid Cells


I think this does well as an answer to your questions!


Thanks for that! This was a very interesting read! I learned very much about how the hippocampus and entorhinal cortex work together to represent location.

However, even after trying to pay close attention to the most interesting and salient information, I’m still unsure how to answer my own questions. The paper was mostly about hippocampal place field locations and place cell remapping, so I still don’t know much about grid cell dynamics or their relationship with sensory input.


One thing I noticed was the huge emphasis on firing rate remapping and rate encoding. Such as this snippet

“the hippocampus can simultaneously convey information related to the position of an animal and to the cues present in the environment. During rate remapping, the integrity of the spatial code is preserved because place fields are stable, but the precise firing rate of neurons varies to encode information not related to the current position of an animal.”

This sounds like an important feature of the brain.


Did you get anything out of figure 3?

Section 3 does call out related papers - I have not looked at them but it may be a good point to dig deeper.

That said - I picked up on that you are looking to see how the senses end up forming these patterns in the EC.

Let me make it clear that nobody knows this.

My twitter feed is full of ads for post-doc positions to research this very question. Whoever does figure this out will prolly score on a Nobel prize for it.


Right I think. I just had these questions after watching Matt’s demo where you can see the grid cells firing when the mouse’s location enters their firing fields. I started to think about how those modules can create the hexagonal grid pattern to encode an environment as large and diverse as the world. I guess I would just need more info about the input to EC.


I just saw this paper today in my twitter feed. It does not directly answer the question of how the senses are encoded in the hippocampus but it does offer more insight into the general topic of coding and retrieval in the medial temporal area:

Reactivated spatial context guides episodic recall - Nora A. Herweg, Ashwini D, Sharan, Michael R.Sperling, Armin Brandt, Andreas Schulze-Bonhage, Michael J. Kahana


And here is another on time representation in lateral EC:


I think I can better articulate my inquiry after thinking this over for a few days.

So, the goal of these grid cells is to uniquely represent each location in an environment by having the hexagonal lattice “cover over” all possible locations of that environment with equally spaced place fields.

Assuming that is an accurate premise, then the question is: how would I initialize these place fields for a given set of grid cells (in modules), assuming this is the very first environment to be learned, so no previous place fields exist? And then how would they grow to represent the environment as it is explored?

AFAICT, you can’t create the grid before you know the extent of the environment, so it has to start from nothing and then be learned through movement, right?

I don’t expect the answer to this to be biologically accurate really, since I highly doubt this is known. Who knows.

I figured I might as well pose this questioning here while I try to hack up my own (probably dumb) solution.

How to generate place fields in grid cells

We have been talking about the spatial component of memory and how that may play out in the hippocampus.
This paper adds some interesting fuel to that fire:

With code:

and the dataset to feed into the process:


We talk about place, direction, edge cells as normal spatial functions of the hippocampus/entorhinal cortex.

For a while now I have been saying that the entire personal experience is coded in this system - not just spatial dimensions. Here is how another major chunk of your experience shows up in this coding: