Dan Levenstein

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...

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

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

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… 🥲)

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.

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?

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

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.

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

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!

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 👇

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

Kauffman Level 3 is when you get the superpowers 👍

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!

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...

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

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

bsky.app/profile/spac...

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

bsky.app/profile/spac...

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

Since it resonated with the audience, I’ll recap my main argument against AGI here. ‘General intelligence’ is like phlogiston, or the aether. It’s an outmoded scientific concept that does not refer to anything real. Any explanatory work it did can be done better by a richer scientific frame. 1/3

On a broader note: Our ANN-based models should generate bold predictions with direct prescriptions for how to test them. This is how I believe #NeuroAI can avoid becoming an isolated, self-referential domain of neuroscience. This is the only way to flywheel NeuroAI into a theory of the brain (2/4)

I liked it better when I thought it was Weird Al 🥲

Living Science: My grandmother’s rug The objects that migrants carry with them are tethers to the lives they have left behind.

Another great essay from Eve Marder’s “Living Science” series, on immigration and the international nature of science 🫶

PFCinema… 😘👌

Super cool!

TFW bsky needs a dislike button…

Old news 😉

If you’re looking for a senior PhD student, I know some neuro and I know some AI…

Congrats 🎈🍾

Using goal-driven deep learning models to understand sensory cortex - Nature Neuroscience Recent computational neuroscience developments have used deep neural networks to model neural responses in higher visual areas. This Perspective describes key algorithmic underpinnings in computer vision and artificial intelligence that have contributed to this progress and outlines how deep networks could drive future improvements in understanding sensory cortical processing.

The exercise of comparing computational models to physiology has quite a long history. E.g a recent approach using task-trained ANNs is often attributed to Yamins/DiCarlo’s work on the visual system.

(rev: www.nature.com/articles/nn....)

Yes! I agree this is going to a big challenge with this kind of model-data comparison… what kind of pretraining do you need to get a good match?

Bet: this flavor of same-stimulus, same-task, compare-behavior, compare-physiology is the future of model testing and theory development in neuroscience.

😳🫣😇

Maybe that’s just my psychology

The funny thing about the brain is that it’s quite adaptive - its operations depend on the situation it finds itself in.

So, do these points of incommensurability reflect different Psychologies (the field) that carve the mind at different joints, or different psychologies (the object of study)?

All cognition is theoretical! Assumptions are inescapable, but some are testable.

“Banger,” my neighbors hear softly through the walls