Lightnews — Scholar-powered news

Joao Barbosa @jbarbosa.org · 56m

Just go across the street and ask him :)

Joao Barbosa @jbarbosa.org · 59m

I don't think that's right. In Mongillo's you can only store in STP what is already in the connectivity/LTP, which in the case of this particular paper was just one of two attractors

1

Joao Barbosa @jbarbosa.org · 1h

just to be clear, Mongillo's model — arguably the model from which all synaptic WM models are based — is also an RNN. I don't think the debate you were alluding to is about RNN vs something else

1

Joao Barbosa @jbarbosa.org · 1h

Yeah, I think that's a reasonable POV. After all, the attractors have to be learned (presumably through LTM, e.g hebbian plasticity).

However, it's the same for synaptic WM: you still need the attractors in the connectivity for the model to work (see Mongillo et al, Science)

1

Joao Barbosa @jbarbosa.org · 1h

sure, if it's low rank :)

4

Reposted by Joao Barbosa

Dawei Bai @daweibai.bsky.social · 23h

Happy to share that our BBS target article has been accepted: “Core Perception”: Re-imagining Precocious Reasoning as Sophisticated Perceiving
With Alon Hafri, @veroniqueizard.bsky.social, @chazfirestone.bsky.social & Brent Strickland
Read it here: doi.org/10.1017/S014...
A short thread [1/5]👇

3 32 78

Joao Barbosa @jbarbosa.org · 23h

There's tricks to chunk items, but there's no debate WM is severely limited. Long term memory is different ofc, virtually infinite

1

Joao Barbosa @jbarbosa.org · 1d

I think I will miss a meeting at 17.30 Nov 6...🙈

1

Joao Barbosa @jbarbosa.org · 1d

Not sure I follow. In networks # of attractors/memories scales with # of neurons. Regarding WM, networks can do MUCH more than brains.

In fact that's *the mystery* worth studying: why can we remember just 4-7 items? Truly mysterious IMO, especially knowing WM capacity strongly correlates with IQ

1

Joao Barbosa @jbarbosa.org · 1d

Why not?

1

Joao Barbosa @jbarbosa.org · 1d

Why? An RNN without STP can do any WM task

1

Joao Barbosa @jbarbosa.org · 1d

I didn't understand, sorry. I agree there are open questions re WM, but synaptic WM doesn't seem to help much.

IMO the only clear advantage (potentially, no evidence AFAIK) is more stability. This comes w/ the cost of across trial interface, known to happen in nature

jbarbosa.org/files/barbos...

jbarbosa.org

Joao Barbosa @jbarbosa.org · 1d

Confusing take. The evolutionary distance between the two species is a beautiful mystery.

The evidence in monkeys isn’t nearly as strong, nor has the same sophisticated experiments, but the mechanism was first proposed in that context.

It's an amazing coincidence? Is it silly? Nature's fault!

Vinny Costa @vincentcostaphd.bsky.social · 1d

Hot take…asking whether mechanisms relevant for flies remembering odors aligns with those underlying working memory in primates is silly.

Mushroom bodies and lateral prefrontal cortex aren’t even close to an example of convergent evolution.

bsky.app/profile/blue...

Alessio Quaresima @bluedebris.bsky.social · 9d

At #BernsteinConference heated debate on WM mechanisms. I love that this is still far to be settled!

trained RNNs from A.Compte aligns with fruitfly direction WM data from K.Nagel: it is persistent.

Large scale recording in macaque V1 in T.Moore Lab: it is not.

#neuroskyence what's your take?

2

Reposted by Joao Barbosa

NYU Center for Data Science @nyudatascience.bsky.social · 1d

CDS Asst. Prof. @neurograce.bsky.social has launched a YouTube channel, “5 Minute Papers on AI for the Planet,” translating climate-AI research into short video explainers, inspired by her course at CDS, “Machine Learning for Climate Change.”

nyudatascience.medium.com/cds-grace-li...

CDS’ Grace Lindsay Launches YouTube Channel on “AI for the Planet”

Grace Lindsay’s new YouTube series turns climate-AI research into five-minute videos for a wide audience.

nyudatascience.medium.com

1 13 29

Reposted by Joao Barbosa

Alireza Modirshanechi @modirshanechi.bsky.social · 1d

After a few days of extensive discussions about curiosity at @rtg2906-curiosity.bsky.social's retreat, I realized—perhaps for the first time—that the word curiosity in Persian (کنجکاوی) literally means "exploring corners"! It's such a vivid way to express curiosity! so amusing to think about it!

1 3 23

Joao Barbosa @jbarbosa.org · 1d

Definitely cool! But far from the kind of evidence that is around for the opposite. The hypothesis is very hard to prove and papers in favor of it essentially start from null results (eg lack of decoding)

2

Reposted by Joao Barbosa

Shahab Bakhtiari @shahabbakht.bsky.social · 1d

Regardless of what explainability/mech interp in AI is actually after, and whether or not they know what they’re searching for, we can confidently say they’re pursuing what systems neuroscience has pursued for decades, with very similar puzzles and confusions.

Mel Andrews @bayesianboy.bsky.social · 1d

What problem is explainability/interpretability research trying to solve in ML, and do you have a favorite paper articulating what that problem is?

2 8 45

Joao Barbosa @jbarbosa.org · 1d

Sure! Or to play a role in shaping activity dynamics. But my point is that, in comparison, there's no real evidence for any of that

1

Joao Barbosa @jbarbosa.org · 1d

Done

1

Joao Barbosa @jbarbosa.org · 1d

There's probably nothing more influential or w/ more explanatory power in neurosci than attractors and more generally computations through population dynamics. In comparison, the level of evidence for fast plasticity being relevant for cognition is close to zero

It's more of a heckler than a debate

Alessio Quaresima @bluedebris.bsky.social · 9d

At #BernsteinConference heated debate on WM mechanisms. I love that this is still far to be settled!

trained RNNs from A.Compte aligns with fruitfly direction WM data from K.Nagel: it is persistent.

Large scale recording in macaque V1 in T.Moore Lab: it is not.

#neuroskyence what's your take?

3 10

Reposted by Joao Barbosa

Nick Brown @steamtraen.eu · 2d

😅

1 13 55

Reposted by Joao Barbosa

samuel mehr @mehr.nz · 3d

postdoc, in an esoteric field, preparing for the job market

2 5 27

Reposted by Joao Barbosa

Helen Barron @helencbarron.bsky.social · 4d

** We have up to TWO funded PhD positions available in our lab!! Apply below to find new ways to enhance memory👇 Pls retweet **

Deadline: 2nd December

1. Cross-species closed-loopTMR: tinyurl.com/bddu4tp6

2. TUS and TMR in humans:
tinyurl.com/jjws5ctj

Happy to chat to interested applicants.

Enhancing memory using cross-species closed-loop Targeted Memory Reactivation | mrcbndu

tinyurl.com

1 42 47

Reposted by Joao Barbosa

Aki Vehtari @avehtari.bsky.social · 4d

I'm looking for a doctoral student with Bayesian background to work on Bayesian workflow and cross-validation (see my publication list users.aalto.fi/~ave/publica... for my recent work) at Aalto University.

Apply through the ELLIS PhD program (dl October 31) ellis.eu/news/ellis-p...

ELLIS PhD Program: Call for Applications 2025

The ELLIS mission is to create a diverse European network that promotes research excellence and advances breakthroughs in AI, as well as a pan-European PhD program to educate the next generation of AI...

ellis.eu

32 40

Reposted by Joao Barbosa

samuel mehr @mehr.nz · 4d

"the quickest way to get a collaborator to return edits on your manuscript is to write down their affiliation wrong" (ancient academic proverb)

2 28 180