Lightnews — Scholar-powered news

Reposted by Sam Gershman

Tomer Ullman @tomerullman.bsky.social · 8h

New preprint!

"Non-commitment in mental imagery is distinct from perceptual inattention, and supports hierarchical scene construction"

(by Li, Hammond, & me)

link: doi.org/10.31234/osf...

-- the title's a bit of a mouthful, but the nice thing is that it's a pretty decent summary

4 16 38

Sam Gershman @gershbrain.bsky.social · 8h

Yeah, I agree. There seems to be a pattern across various "adversarial" papers where they feel the need to take down the orthodoxy. I think it's more often "yes and" rather than "no but".

1 1 8

Sam Gershman @gershbrain.bsky.social · 9h

I get a dopamine burst every time I see a new idea about dopamine.

3

Sam Gershman @gershbrain.bsky.social · 10h

To be clear, their "adaptive gain" hypothesis allows dopamine to have a kind of auxiliary role in learning by controlling action repetition/vigor, which in turn changes downstream learning signals.

1 18

Sam Gershman @gershbrain.bsky.social · 10h

I'm not yet convinced that dopamine has no role in learning, but I think the authors make a persuasive case that dopamine has a role in action control that goes beyond learning.

1 15

Sam Gershman @gershbrain.bsky.social · 10h

Really interesting work by Bakhurin and colleagues challenging the reward prediction error hypothesis of dopamine:
www.nature.com/articles/s41...
I love this figure which both echoes and undermines the famous figure from Schultz et al. (1997).

3 40 120

Sam Gershman @gershbrain.bsky.social · 1d

What you really need are the howlers from Harry Potter.

a group of people standing around a red box with the words `` how dare you '' written on it .

ALT: a group of people standing around a red box with the words `` how dare you '' written on it .

media.tenor.com

1 2

Sam Gershman @gershbrain.bsky.social · 4d

My lab started doing molecular biology experiments precisely to try to answer these questions. This is actually why I started getting interested in these phenomena again (we have evidence for anti-contiguity effects in single cells). Stay tuned!

3

Sam Gershman @gershbrain.bsky.social · 4d

Gallistel has developed multiple models to explain the effects of timing on conditioning, although these are not really mechanistic. Here's a recent one:
brill.com/view/journal...
And here is my more mechanistic take:
gershmanlab.com/pubs/Gershma...

gershmanlab.com

1 3

Sam Gershman @gershbrain.bsky.social · 4d

I agree with the general logic of this point, but then you would have to explain why the minimum processing time varies by orders of magnitude across preparations. In the flavor aversion example, it's extremely unlikely that animals need to taste the flavor for an hour.

1

Sam Gershman @gershbrain.bsky.social · 4d

Hm, I think you are using a different definition of contiguity from the one that appears (for example) in Rescorla's paper, which discusses the interval between CS onset and US onset. I think that is the most relevant definition for associative theory.

1

Sam Gershman @gershbrain.bsky.social · 4d

I don't think so, because this implies that if you hold T fixed to a small value and reduce I sufficiently, then you should get faster learning, but in fact this slows learning (i.e., the effect of I is typically monotonic). Ths figure is also from Rescorla's 1988 paper, using Gibbon's data.

1 3

Sam Gershman @gershbrain.bsky.social · 4d

There's this paper on the definition of learning:
link.springer.com/article/10.3...
but that doesn't really provide a taxonomy. You can find taxonomies in some specific domains, but nothing super broad as far as I know.

link.springer.com

1 2 6

Sam Gershman @gershbrain.bsky.social · 4d

I think for that account to have legs, you'd have to be more specific about what you mean by attentional mechanism.

1

Sam Gershman @gershbrain.bsky.social · 4d

Great question. I don't know! I don't think it's reducible to some kind of simple biophysical constraint, because it seems to operate over multiple timescales. It could be a kind of causal prior, perhaps. To get at that, you'd want to see if you could manipulate it through training.

2

Sam Gershman @gershbrain.bsky.social · 4d

I'm teaching a class on theories of learning, and we had a great discussion about what makes associative learning truly associative, and I pointed to these data as evidence that what we call associative learning isn't actually associative. Most students had never seen the data before.

3 10

Sam Gershman @gershbrain.bsky.social · 4d

The anti-contiguity effect violates Gallistel & Gibbon's rate estimation theory, which is notable given that its evident in Gibbon's own data (the top middle plot in Rescorla's figure). I'm not aware of any theory that explains this (but please someone correct me if I'm wrong!).

2 6

Sam Gershman @gershbrain.bsky.social · 4d

There's another, often less appreciated violation of contiguity, which was nicely documented in this figure from Rescorla (1988). For very short interstimulus intervals, the conditioned response and learning speed *increase* with the interval. This is an anti-contiguity effect!

4 1 13

Sam Gershman @gershbrain.bsky.social · 4d

This is a great idea except for the fact that it's wrong. You can obtain the same learning speed across dramatically different interstimulus intervals (T) as long as the intertrial interval (I) is kept in a fixed ratio with the interstimulus interval (Gallistel & Gibbon, 2000).

1 4 25

Sam Gershman @gershbrain.bsky.social · 4d

Everyone knows that temporal contiguity is important for associative learning. As the interval between a cue (e.g., a light) and an outcome (e.g., shock) gets longer, the conditioned response (e.g., freezing to the tone) is acquired less quickly.

1 7 45

Sam Gershman @gershbrain.bsky.social · 6d

"War on unicorns"
or
"War by unicorns"
?

2

Reposted by Sam Gershman

Ashley Thomas @ashleyjthomas.bsky.social · 8d

Interested in understanding how young humans think about social relationships? I am reading PhD applications this year! **Please note**, that Harvard now requires the GRE. More information here: www.ashleyjthomas.com/workwithme

WANT TO WORK WITH ME? | Mysite

www.ashleyjthomas.com

1 10 15

Reposted by Sam Gershman

Merriam-Webster @merriam-webster.com · 13d

We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.

660 8.3K 25K

Reposted by Sam Gershman

Mike Frank @mcxfrank.bsky.social · 14d

Ever wonder how habituation works? Here's our attempt to understand:

A stimulus-computable rational model of visual habituation in infants and adults doi.org/10.7554/eLif...

This is the thesis of two wonderful students: @anjiecao.bsky.social @galraz.bsky.social, w/ @rebeccasaxe.bsky.social

conceptual schema for different habituation models

1 27 70

Sam Gershman @gershbrain.bsky.social · 15d

I suggested that play is an example where the metric maximized by adults in the ARC games (task completion) is precisely the metric minimized by children during play: the ideal imaginative play is never completed. "Success" is failure. Learning "efficiency" in this setting is irrelevant.

1 1 9