Dan Levenstein

Already having #cosyne2026 FOMO and it hasn’t even started yet 😿

IMO there’s more theory in neuroscience than we give it credit for, but it doesn’t percolate through the field as much as you would hope/need for broader understanding.

I want to write a fun little post on what we've learned in neuroscience in the last 20 years. What are the most interesting results you can think of? Biggest trends?

Neural geometry in the human hippocampus enables generalization across spatial position and gaze Hippocampal neurons track positions of self, others, and gaze direction. However, it is unclear how their respective neural codes differ enough to avoid confusion while allowing for abstraction. We re...

Compositional representation of self, others, and gaze direction in *human* hippocampus - super cool.

arxiv.org/abs/2603.04747

Academics Need to Wake Up on AI Ten theses for folks who haven't noticed the ground shifting under their feet

I'm not gonna touch most of this but two things:
1. The apprenticeship model of science was never supposed to be about the PI's workflow. The point isn't to have research assistants to help to publish stuff faster. Training the next generation researcher *is* the point. Because we care. Allegedly.

science is very very international which is very very good for science and very very good for the world

it's worth remembering that this + the nature of networks means that when bad things happen in any single part of the world, all scientists are only 1-2 degrees separated from it

These reviews on neural manifolds are super relevant to today’s world of large-scale population recordings. How do we link circuits, population geometry, dynamics, and function?

“A neural manifold view of the brain” from Perich, Narain, and Gallego 2025

www.nature.com/articles/s41...

(1/2)

I’m building a foundational reading list for our lab (systems & circuit neuroscience, compneuro, modeling, neuromodulators, population coding etc.).

I’d like to crowdsource recommendations.

Which review(s) would you consider mandatory reading for the next generation of researchers?

Papez!

Yes! And hopefully that grows 🌱. I think the biggest difference is just the sheer volume of researchers…

🫣

One way to see this is the way “feedback” is often used. Some parts of the field use feedback as coming from other areas. Sometimes ppl overload it to mean local input too.

I think it’s more useful to name inputs as coming from one of three sources- feedforward, feedback, or recurrent.

TBH the thing that is most sad about the loss of ~2015-2021 era science twitter is the loss of a public square where grad students can see, and participate in, the diversity of expert opinions in the field. It really shaped how I see science.

TBH these are some of the most creative times in research and if Claude et al bring them back I’d call that a win. 🏓🤺

Liftoff: New lab alerts Learn about early-career scientists starting their own labs.

In this month’s “Liftoff,” @qlu.bsky.social talks about how he will use a “slide stack” to keep up with his students’ projects, and @dlevenstein.bsky.social shares how he will continue a time-honored tradition he learned in György Buzsáki’s lab.

By @franciscorr25.bsky.social

bit.ly/4apOZm9

🦆🦆🦆😍

one of the active states is definitely bursting ;) (check e.g. CA1 during NREM... biggest density of burst-timescale ISIs).

most regions/brain states have multiple AS modes, there are others that are slower than bursting (e.g. in the 5Hz-100Hz range). bursts are one animal in the modal zoo 🦁🐯🐻😱

Yes absolutely! Love Chai's paper. We actually met at that 2020 Cosyne and hit it off. I think it's highly relevant to GS spiking ;)

I shall not attempt further to define the kinds of material I understand to be embraced within that shorthand description (“representation”), and perhaps I could never succeed in intelligibly doing so. But I know it when I see it and also you better not tax me unless I have it.

Thanks Julie! Hope everything is going well down at Princeton :)

Just wait for whiteboard meetings with students days 👍

IMO there's quite a bit of follow up work to do re: AS/GS modes on a spike-by-spike basis (notice everything in the paper is statistics over conditional ISI distributions), and how they relate between cells. We tried a few different approaches and I wasn't satisfied with any of them/ran out of gas⛽️⌛️

Thanks Jay! We tried to answer this question and from what I remember (it's been a while), it looked like yes - GS spikes are more correlated than you would expect from asynchronous spiking. However, I remember it being challenging to say anything confidently and this didn't make it to the paper...

What about GS mode? We think the brain uses GS mode to maintain its own dynamics. A constant buzz of GS spikes keeps cells near-threshold, and maintains a diverse population of spike sequences to select from and reinforce when associated with salient behaviors or stimuli.

Why might the brain work like this? Our idea is that having discrete modes supports functional multiplexing. Spiking at specific timescales could resonate with different downstream readers and engage intracellular/synaptic mechanisms with different functional implications.

As a proof of principle, we built a spiking model in which cells used iSTDP to balance at a cell-specific rate. In the standard balanced condition, cells only had GS mode. When we adjusted network parms to other activity regimes, AS modes emerged but notably, GS was preserved.

So what do we think is going on here? TL;DR: AS modes reflect a cell's engagement with various region/state-specific activity patterns. GS mode reflects E/I balanced inputs, and each neuron has a different balance point that sets its GS rate.

(see also: pubmed.ncbi.nlm.nih.gov/36086892/)

On the other hand, neurons spent the most time in GS mode and their mean rate mainly reflected of GS rate (rather than AS rate/weight). GS rate was conserved across WAKE and NREM, suggesting it's characteristic feature of a cell.

We reproduced this observation in the thalamic head direction system, and even found that a “modal” encoding model, in which probability of AS mode varied with HD, fit HD cells better than a continuous rate model.