Testing Assumptions Underlying a Unified Theory for the Origin of Grid Cells

Authors: Rylan Schaeffer, Mikail Khona, Adrian Bertagnoli, Sanmi Koyejo, Ila Rani Fiete

Venue: NeurIPS 2023 Workshops:

• Symmetry and Geometry in Neural Representations
• Unifying Representations in Neural Models
• AI for Science

Quick Links

• Tweeprint
• Paper
• Poster
• Recorded Talk
• Slides

Summary

Announcing #2 in our #NeurIPS2023 workshop & conference papers series (2/10)!!

🔎🧠Testing Assumptions Underlying a Unified Theory for the Origin of Grid Cells🔎🧠

w @KhonaMikail @FieteGroup @sanmikoyejo & Adrian Bertagnoli

Appearing @neur_reps @unireps @AI_for_Science

https://openreview.net/forum?id=vYixJUwAD4

🧵👇

1/8

Grid cells are a Nobel-prize winning neural representation found in the mammalian brain, that play a fundamental role in spatial navigation

Where do these representations come from?

2/8

One leading theory in #NeuroAI is the so-called “Unified theory for the origin of grid cells” by @SuryaGanguli Ben Sorscher & Gabe Mel

The Unified Theory posits that grid cells arise due to predicting place cells, another type of neural representation

3/8

We identify 2 critical mathematical assumptions that the Unified Theory rests upon:

1) Place cells, as a population, must be translationally invariant

2) Place cells, individually, must have center-surround tuning

Is either assumption biologically valid?

Let’s test !

4/8

Using 320 place cell recording sessions, we propose 2 ways to quantify whether place cells are translationally invariant and also qualitatively visualize their spatial autocorrelations

Conclusion: Place cells unlikely to be translationally invariant ❌

5/8

Extracellularly, place cells don’t exhibit center-surround tuning.

We dig through the literature to discover place cell subthreshold voltages also do not exhibit center surround tuning (right)

Conclusion: Place cells unlikely to have center surround tuning ❌

6/8

Together, these cast doubt on Unified Theory’s relevance to bio grid cells @unireps: https://openreview.net/forum?id=vYixJUwAD4

@neur_reps: https://openreview.net/forum?id=CwJIpWzgDP

@AI_for_Science: https://openreview.net/forum?id=vYixJUwAD4

Work done @ @mitbrainandcog @mcgovernmit @stai_research @StanfordData @StanfordAILab

7/8

What are your thoughts on these findings? Join the conversation at #NeurIPS2023 !

8/8