"The inevitability and superfluousness of cell types in spatial cognition" w Ken Luo @robmok.bsky.social Whether place, border, head direction, Jennifer Aniston, or whatever cells, are we fooling ourselves? Are these intuitive findings scientific discoveries? 1/6 www.biorxiv.org/content/10.1... - ThreadSky

profdata.bsky.social • 405 days ago

"The inevitability and superfluousness of cell types in spatial cognition" w Ken Luo @robmok.bsky.social Whether place, border, head direction, Jennifer Aniston, or whatever cells, are we fooling ourselves? Are these intuitive findings scientific discoveries? 1/6 https://www.biorxiv.org/content/10.1101/2024.01.10.575026v1

Comments

profdata.bsky.social•405 days ago

We find spatial cell types are (1) INEVITABLE: they occur in sufficiently complex (incl. random) networks and (2) SUPERFLUOUS: decoding is fine without these units. We tested an agent in a VR rodent enclosure with viewpoints as inputs to a deep net for object recognition. 2/6

profdata.bsky.social•405 days ago

Object recognition models are non-spatial. Indeed, they are trained to be translation invariant. Nevertheless, direction and position are easily decoded from these models, including visual transformers and convolution nets with random weights. 3/6

profdata.bsky.social•405 days ago

All your favorite cell types can be found in these models, despite these networks performing decidedly non-spatial functions. Units with these response properties appear INEVITABLE in complex networks, including random nets. 4/6

profdata.bsky.social•405 days ago

Further, excluding these "spatial" units has little affect on decoding performance. They are SUPERFLUOUS. 5/6

profdata.bsky.social•405 days ago

Intuitive cell types should be relegated to history. They arise in networks performing unrelated functions. You might get a prize for finding them, but it means little scientifically. Scientists, incl. modelers, should be weary of their preconceptions of how the brain works. 6/6

hugospiers.bsky.social•405 days ago

Love the shots fired here Brad! Too few argumentative articles. HD cells are the ones I’m most curious about. Have you had any open disagrees yet?

jbarbosa.org•405 days ago

Very cool! But the real mystery is grid cells, aren't they?

profdata.bsky.social•405 days ago

thanks! we already solved that, haha, https://doi.org/10.1038/s41467-019-13760-8 , they are fictional as well :-) I can check but I think we found units even in the present work that passed the grid score criteria.

hugospiers.bsky.social•405 days ago

What cells are predicted that oddly don’t seem to have been reported(yet)? Eg cells that separate two sides of room with some nice gradient?

charan-neuro.bsky.social•405 days ago

Preach! I would add that functionally defined cell types are invariably overfitted to the narrow experimental contexts that are used to define them. The way the question is posed will constrain what answers you will see. However, I am guessing the editors of Cell will not be pleased...

profdata.bsky.social•405 days ago

100% This paper with @robmok.bsky.social basically argues grid cell response is an artifact of the experiment task, https://doi.org/10.1038/s41467-019-13760-8 . I don't think people like to hear this. We'll see how this goes in review!

neurostats.org•49 days ago

📌

charan-neuro.bsky.social•48 days ago

I'd add something. Studies of place cells are done on animals with few meaningful experiences, which is like putting an infant in an IKEA, not like humans navigating the real world. Idk about place cells in animals that have a lot of experiences in similar environments (i.e., what humans do).

pessoabrain.bsky.social•405 days ago

This looks fascinating and will take a careful look soon. However deep neural networks are a rather impoverished model of the brain... But as a existence proof it provides valuable information ofc.

profdata.bsky.social•405 days ago

Thanks! Yeah, I think impoverished here is a strength because it helps the argument that basically any complex network will give you these cell types, even random weights.

pessoabrain.bsky.social•405 days ago

good point! :-)

jaquent.bsky.social•405 days ago

This reminds me of the fact that it has been previously thought "place cells" only code small areas and exists only in the dorsal HC but when they tested larger spaces they found them in the ventral HC as well.

Anyway, this paper is definitely on my reading list. I love a good old controversy.

jkubie.bsky.social•396 days ago

... and I'm not sure what decoding tells you. From the firing of the population of optic nerve neurons you could likely decode the visual scene and objects in it. What does that say about visual cortex?

jkubie.bsky.social•396 days ago

Also, let's imagine there are place cells and interneurons that do local processing to sharpen place cells, or something like that. For example, "theta cells", that are presumed to be inhibitory interneurons, do have place fields, but have consistent spatial patterns. Likely these alone could do it.

jkubie.bsky.social•396 days ago

As I see it, the important question is not whether all of the cells in the network have spatial signals, etc, but what is the collective signal of the output cells, in that the output cells determine how the information is used downstream.

profdata.bsky.social•395 days ago

I am not appreciating how your statements are in conflict with the fundamental message of the paper.

Comments

Posting Rules

Reply