Computer scientists prove why bigger neural networks do better

Re:Computer scientists prove why bigger neural networks do b... by mattkrause on Thursday February 10, 2022 @04:00PM (#30292431)

As a counterpoint, rats without a cortex can do...basically everything normal rats can do--except trim their toenails. The classic reference for this is Whitslaw's 1990 chapter "The decorticate rat".

This thread has links to a copy, plus a bunch of related studies in humans and animals. https://twitter.com/markdhumphries/status/107105276276554137...

[ Parent | on HN ]

Re:Computer scientists prove why bigger neural networks do b... by heavenlyblue on Thursday February 10, 2022 @07:19PM (#30294943)

Have you read the PDF given?

The whole nail part is basically a single sentence in the paper.

For example, decorticate rats are unable to escape narrow alleyways because they can not turn around due to their tonsils touching the walls and them being unable to ignore that feeling.

Another example is that they take a few seconds vs (!) 5 minutes to groom themselves on average.

[ Parent | on HN ]

Re:Computer scientists prove why bigger neural networks do b... by mattkrause on Thursday February 10, 2022 @07:59PM (#30295349)

Yes! I got interested in this when my colleague worked with a person who had an entire hemisphere resected as a teenage.

FWIW, the nail thing is a bit of a neuroscience meme. I heard--and stole--this quip from multiple people in several different situations. There's also a really striking figure in that chapter (p. 7 or 8).

No argument that the rats' behaviors are affected. I suppose whether you find the slowness of their grooming expected (because of brain damage) or impressive (because it happens at all) is a matter of taste. Glass^W Skull half-empty or half-full, if you will.

[ Parent | on HN ]

Re:Computer scientists prove why bigger neural networks do b... by heavenlyblue on Thursday February 10, 2022 @08:15PM (#30295488)

> No argument that the rats' behaviors are affected. I suppose whether you find the slowness of their grooming expected

They are not slow, they just stop grooming themselves well enough.

[ Parent | on HN ]

Re:Computer scientists prove why bigger neural networks do b... by gwern on Thursday February 10, 2022 @08:09PM (#30295439)

Stick that rat outside in a wild rat colony, and we'll see how well "everything normal rats can do except trim their toenails" works out.

[ Parent | on HN ]

Re:Computer scientists prove why bigger neural networks do b... by mattkrause on Thursday February 10, 2022 @08:31PM (#30295610)

The cool thing about Whitslaw's work is that it focused on natural behaviors (rather than like...a rotorod test).

I don't think he released them into the wild (would be a tough experiment with 80s tech), but there are a bunch of studies of their interactions with conspecifics. They can mate[0], though less successfully than controls, but playfight a bit better than they do[1].

[0] https://psycnet.apa.org/record/1983-29790-001 [1] https://doi.org/10.1159/000114124

[ Parent | on HN ]