Lightnews — Scholar-powered news

Reposted by Dan Levenstein

Alex Williams @itsneuronal.bsky.social · 14h

It's possible to get first order approximate understanding of RNNs performing relatively complex tasks.

[1906.10720] Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamics share.google/ElT686dgAUIk...

But some tasks are harder than others 🤷‍♂️

Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamics

Recurrent neural networks (RNNs) are a widely used tool for modeling sequential data, yet they are often treated as inscrutable black boxes. Given a trained recurrent network, we would like to reverse...

share.google

1 2

Reposted by Dan Levenstein

Sam Gershman @gershbrain.bsky.social · 16h

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 9

Reposted by Dan Levenstein

Tom Gauld @tomgauld.bsky.social · 2d

My cartoon for this week’s @newscientist.com

A comic strip in six panels. The panels inerlock like a jigsaw with knobs and sockets. Two stick figures walk right to left through all the panels.

Panel 1: Person 1 says "I hear you're working on a new theory."
Panel 2 & 3: Person 2 says: "Yes, I'm calling it the jigsaw model. I think it could change our whole View of reality."
Panel 4: Person 1 says "When Will you publish?:
Panel 5: Person 2 says "But a couple of things are still bothering me." (this piece/panel is a bit squint)
Panel 6: This piece is rotated ninety degees and fits to the others very badly. Person 2 is saying "Soon. I feel like I'm really close"
A second look at the whole comic shows that the last two panels read in the wrong order because the jigsaw has been put together incorrectly.

16 390 1.4K

Reposted by Dan Levenstein

Tatiana Engel @engeltatiana.bsky.social · 3d

We apply our model to survey the spiking irregularity across cortical areas and find that Poisson irregularity is a rare exception, not a rule. Our results show the need to include non-Poisson spiking in inferring neural dynamics from single trials.

2 9

Dan Levenstein @dlevenstein.bsky.social · 3d

Language is a great example - not necessary for intelligence to emerge, but might be enough to bootstrap. Once you’re using language data you’re not building intelligence de novo.

I’d bet social interactions are the opposite - you need them for intelligence to emerge but not to bootstrap.

1 4

Dan Levenstein @dlevenstein.bsky.social · 3d

Always find it an interesting question which ingredients are necessary for intelligence to emerge from scratch vs which are necessary if you only want to bootstrap from an existing intelligent system…

Ricard Solé @ricardsole.bsky.social · 3d

Moreover, AGI might also require social interactions to become a reality, where cultural evolution (and extended mind) play a major role. It is not by chance that brain and culture evolved together, allowing complex minds to emerge @anilseth.bsky.social @mitibennett.bsky.social

1 2 12

Reposted by Dan Levenstein

Antonino Greco @agreco.bsky.social · 3d

here the trick is to unpack the word *understanding*

it’s not about what each neuron does, but the rules shaping the behaviour of biological systems!

"rules for development and learning in brains may be far easier to understand than their resulting properties"

👇👌
arxiv.org/abs/1907.06374

1 4 28

Dan Levenstein @dlevenstein.bsky.social · 4d

It is possible that really appreciating this tweet requires having gone to college in the states around 2010 😅

4

Dan Levenstein @dlevenstein.bsky.social · 4d

I’m at the restaurant and they’re playing my song and can you imagine anyone making a song right now called “Party in the USA”?

1 8

Reposted by Dan Levenstein

Grace Lindsay @neurograce.bsky.social · 4d

That's what we're saying people should try to do here: www.sciencedirect.com/science/arti...

Testing methods of neural systems understanding

Neuroscientists apply a range of analysis tools to recorded neural activity in order to glean insights into how neural circuits drive behavior in orga…

www.sciencedirect.com

1 3 16

Dan Levenstein @dlevenstein.bsky.social · 4d

Turns out @macshine.bsky.social scooped our title….

www.biorxiv.org/content/10.1...

Can Systems Neuroscientists Understand an Artificial Neural Network?

Network neuroscience has catalysed crucial insights into the systems-level organisation of the brain, however the lack of a ‘ground truth’ inherently limits direct interpretation. In parallel, deep le...

www.biorxiv.org

1 9

Dan Levenstein @dlevenstein.bsky.social · 4d

The constraints and self-regulation are the special sauce. I like the point that understanding self-regulation could be a secret back door (“leverage”) to understanding computation (or maybe not needing to… 🥲)

1 6

Dan Levenstein @dlevenstein.bsky.social · 4d

A human-comprehensible story about how the pattern of activations lead to a network’s competencies in real-world tasks, and how they come to do so with learning. Which we can back up with predictions and perturbations. TL;DR the dream of systems neuroscience.

3

Dan Levenstein @dlevenstein.bsky.social · 4d

Because a deep RNN is a much simpler system than the brain, that also operates through the parallel/distributed activity of connected input/output units, where the efficacy of connections plays a key role in its operation. If we can’t understand that how can we hope to understand the brain?

1 9

Dan Levenstein @dlevenstein.bsky.social · 4d

So I get that a Neuroscientist Couldn’t Understand a Microprocessor, and TBH I’m ok with that. But could a neuroscientist understand a deep RNN? Because that seems like a more pressing issue.

*assuming you think the brain operates through the parallel activity of many connected input/output units

14 4 50

Reposted by Dan Levenstein

Shahab Bakhtiari @shahabbakht.bsky.social · 6d

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 · 6d

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

3 8 46

Dan Levenstein @dlevenstein.bsky.social · 6d

And if you’re looking for a postdoc not a faculty position, we have those too 😉

Dan Levenstein @dlevenstein.bsky.social · 6d

Come do a postdoc at the Wu Tsai Institute!

WTI fellows have freedom to work with anyone at the institute, and preference is given to applicants who want to work on interdisciplinary projects with multiple faculty mentors.

If you’re interested to work with me, please reach out!

Wu Tsai Institute | Yale University @wutsaiyale.bsky.social · 8d

📣 Calling experimental, computational, or theoretical researchers!

WTI's Postdoc Fellowships application is now open, offering a competitive salary, structured mentorship, world-class facilities + more: wti.yale.edu/initiatives/...

Apply by November 10: apply.interfolio.com/174525

#KnowTogether

1

Dan Levenstein @dlevenstein.bsky.social · 6d

Come do a postdoc at the Wu Tsai Institute!

WTI fellows have freedom to work with anyone at the institute, and preference is given to applicants who want to work on interdisciplinary projects with multiple faculty mentors.

If you’re interested to work with me, please reach out!

Wu Tsai Institute | Yale University @wutsaiyale.bsky.social · 8d

📣 Calling experimental, computational, or theoretical researchers!

WTI's Postdoc Fellowships application is now open, offering a competitive salary, structured mentorship, world-class facilities + more: wti.yale.edu/initiatives/...

Apply by November 10: apply.interfolio.com/174525

#KnowTogether

10 19

Dan Levenstein @dlevenstein.bsky.social · 6d

The Wu Tsai Institute at Yale is hiring another faculty member in neurocomputation. Come work with us in a growing community at the interface of neuroscience and AI!

More info below 👇

Wu Tsai Institute | Yale University @wutsaiyale.bsky.social · 11d

📣 WTI is hiring faculty positions! Are you interested in advancing our understanding of the brain + how it gives rise to cognition?

Two calls are open:

Open-rank search, Neurocomputation, deadline: 12.1.25
Senior search, Neurodevelopment, rolling review

🔗 wti.yale.edu/opportunities

#KnowTogether

1 10 19

Dan Levenstein @dlevenstein.bsky.social · 10d

Sounds like a…. bitter pill for them to swallow, eh? 😅🥁

1 3

Dan Levenstein @dlevenstein.bsky.social · 11d

Kauffman Level 3 is when you get the superpowers 👍

1 2

Dan Levenstein @dlevenstein.bsky.social · 11d

Ty! 🙏🙏🙏 We’ll have an updated preprint soon - with non-spatial representations (“splitter”, “lap”, etc cells), an orthogonalized manifold, spatial cell “type” quantification, and sparse-lognormal connectivity.

Also a package+tutorial so you can easily train sequential pRNNs in your own environment!

Reposted by Dan Levenstein

Ching-Lung Hsu @hiallen72.bsky.social · 13d

Beautiful work, as I'm thinking about the algorithmic functions of hippocampal theta!
doi.org/10.1101/2024...

Sequential predictive learning is a unifying theory for hippocampal representation and replay

The mammalian hippocampus contains a cognitive map that represents an animal’s position in the environment [1][1] and generates offline “replay” [2][2],[3][3] for the purposes of recall [4][4], planni...

doi.org

1 1 11

Dan Levenstein @dlevenstein.bsky.social · 12d

📌 This feed to see what the people who like the same things you like like. 🫧

bsky.app/profile/spac...

spacecowboy @spacecowboy17.bsky.social · Aug 3

Welcome to the ✨For You✨ feed!

It finds people who liked the same posts as you, and shows you what else they've liked recently.

📌 Pin to add it to your top bar
❤️ Like the feed and repost to spread the goodness

1

Dan Levenstein @dlevenstein.bsky.social · 12d

New feed based on your co-likers’ likes just dropped

bsky.app/profile/spac...

spacecowboy @spacecowboy17.bsky.social · Aug 3

Welcome to the ✨For You✨ feed!

It finds people who liked the same posts as you, and shows you what else they've liked recently.

📌 Pin to add it to your top bar
❤️ Like the feed and repost to spread the goodness

1