How does the brain decide? 🧠 Our new @nature.com paper shows that neural activity switches from an 'evidence gathering' to a 'commitment' state at a precise moment we call nTc. After nTc, new evidence is ignored, revealing a neural marker for the instant when the mind is made up. rdcu.be/eGUrv

"if something doesn't tell you the answer, it doesn't tell you anything" spot the issue there.

The misleading manifold? The current debate (decoding vs causal relevance) and a toy example I gave in the thread below got me thinking about a related issue: how decoding may reflect structure more than function. 🧵 1/5

The real question is: what should children and students learn? Math, spelling, reading, and especially reasoning! The goal of education isn't to spit out as much text as possible.

The two main pillars of my research vision: Control Theory, and Reverse Engineering.

a doodle from dissertation #happylaborday

A Healthy Academia is a Low Pass Filter. Not aspiring to track the hype on an hour-by-hour basis is kind of our job.

All of NIH funding to Northwestern University is frozen. This pause includes noncompeting approved funding, new and competing grants with fundable scores. No reimbursements for money already spent have been received since March. This situation is rarely reported so please Please get the word out!

My initial thoughts on the NIH + FDA refocus on human subjects research medium.com/neuroenginee...

So, how are feeling about the NIH prioritizing human subjects research? www.nih.gov/news-events/...

I believe this (friendly piece) aligns with your question about the absurdity of reducing strong emotions & compulsions to happenings in the brain. Bonus: it captures a diversity of researcher perspectives about the best ways to think about it. www.thetransmitter.org/the-big-pict...

I'd put these on the NeuroAI vision board: @tyrellturing.bsky.social's Deep learning framework www.nature.com/articles/s41... @tonyzador.bsky.social's Next-gen AI through neuroAI www.nature.com/articles/s41... @adriendoerig.bsky.social's Neuroconnectionist framework www.nature.com/articles/s41...

Let it be known that I too will accept $2 billion in funding on the promise that I will not release anything until I have created God. pivot-to-ai.com/2025/03/06/i...

The main NeuroAI vision is anchored in the "big data" approach of LLM/NLP - even as it's becoming increasingly unclear whether LLM/NLP even works as anything but a language mimic. Mimicry is not what neuroscience was every aiming for...

been saying it for a while but anything about "Ai replacing employees" is a marketing exercise, and if I'm honest I lump in most AI safety conversations with it, because they never seem concerned with the actual harms. It's all so cynical.

I’m hearing that some MOSAIC fellows are not only losing their grants (and for some in the postdoc phase their livelihoods), but also that their LRPs are now being cancelled. To have this population of scientists be targeted in such a manner should be unacceptable to everyone 😤

My PhD student's NSF grant was terminated today. Her dissertation investigates why air pollution monitoring, metrics, and regulations don't properly capture the harms experienced by marginalized communities--esp in the most polluted Detroit neighborhoods. She and I are happy to talk with press etc.

The brain is a massive tangle, so it's always welcome when we can simplify the challenge of understanding it. One of the best steps to simplify: anchor in *patients* and *physician* dynamics.

How would you answer the question: "What do Physicians Do?"

INCREDIBLE opinion by Wilkinson refusing to stay Judge Xinis's orders in the Abrego Garcia case, condemning DOJ's arguments as "shocking not only to judges, but to the intuitive sense of liberty that Americans far removed from courthouses still hold dear." storage.courtlistener.com/recap/gov.us...

How does the brain integrate information from new experiences while preserving established memories? Have a look at our latest preprint, "Experience reorganizes content-specific memory traces in macaques" 🧵 🧪🧠🤖👩‍🔬 www.biorxiv.org/content/10.1...

I want to coin the term "meta-medicine" as an alternative to "healthcare economics" since the latter is so tightly linked to the US-specific implementation of healthcare.

I love working with applied mathematicians.

"Intuition" is a fascinating concept because, as a core pillar, intelligence involves understanding the best way to "rotate" the maps we've estimated into novel scenarios. Intuition plays a role (non-robust) in that rotation.

Not a mistake, a signal.

"Dreamer learns a model of the environment and improves its behaviour by imagining future scenarios." Now we are getting somewhere. @virati.bsky.social @kalletimperi.bsky.social 🧪 bsky.app/profile/natu...

When we use TMS to treat depression, we think it alters neural activity in a deep brain structure known as the subgenual anterior cingulate (sgACC). But this is hard to test directly, until our preprint with @coreykeller.bsky.social, @neuro-engineer.bsky.social, Nick Trapp, and Aaron Boes (UIowa) 1/

In the neuroscience space: AI is a toolset, like "power tools". We need *frameworks* that inform how we use these tools. AI does not provide frameworks - only fields like Control Theory do.

I've been walking around with the working understanding that DL = inefficient function approximation and it's honestly served me very, very well.

If you are at #cosyne2025 come check out my poster this afternoon! We demonstrate a case where objective function design can systematically improve neural predictivity in deep networks. [2-036] Contrastive-Equivariant Self-Supervised Learning Improves Alignment with Primate Visual Area IT

F mechanisms are student and postdoc fellowships. This is the US choking off funding for training the next generation of talent. Where is the university leadership response?

Trump admin doesn't support replications to improve science ofc but wants to use them to bring long discredited ideas, flawed research questions and methods back into circulation. We should have collectively resisted the weaponization of replications as some magical demarcation criterion.

Many law professors really discount the importance of legal storytelling but right now it's more critical than ever. Documenting things and telling stories, esp. in a way that's accessible to folks who aren't legally trained, is one of the better ways we have to influence public opinion.

ICYMI: Beyond Mechanism—Extending Our Concepts of Causation in Neuroscience onlinelibrary.wiley.com/doi/10.1111/...

Code is not documentation. Period. But good code can make it easier to work with in the absence of documentation.

See our new preprint focused on causal thinking for data analysis in neuroscience #causal #neuroscience #neuroimaging #datascience #poldracklab with Brian Caffo, Maya Mathur, and @russpoldrack.bsky.social (1/4) arxiv.org/abs/2503.10710

New blog-post on the logic of mechanistic modeling. open.substack.com/pub/kording/...

Why can't gradients be the solution to learning and more? To have gradients you need a cost function To have cost function you need to know *all* variables that matter & functional form & have function stable across time This doesn't work for adaptive&autonomous biological systems => no gradients

A handy counterexample to this: node perturbation (journals.aps.org/prl/abstract...). Totally local learning rule that behaves *as if* following a gradient. All it needs is reward and local correlations.

My case for my academic research lab is simple and, as far as I can tell, unique: let's bring math in to strengthen and account for intuition(s), instead of decimating intuition for daisy-chained guidelines.

A friend drove two hours today for her breast cancer chemotherapy only to learn it was canceled due to NIH cuts. Clinical trial ended midstream. No treatment, no restart, no plan, no information. Her bro-in-law's chemo was also canceled.

Friendly reminder to everyone working in *clinical* neuromodulation - reductionist rigor and pedantry did not bring us to where we are today.

Your model needs to be expressive enough to capture the main processes of interest, but not so expressive that the data requirements explode.

There's a tension in medicine between folks that think rigor = authoritative and rigor = math checks out.

It's often easier for me to describe my research vision by the methods I want to use: toy-dynamical models in networks, model predictive control, nonlinear + optimal control, SSM AI, applied math + expertise-derived models to a priori constrain solution spaces, etc.

Most people probably aren’t tracking this but Senate Republicans just started an all-night voting marathon that will allow them to gut Medicaid and increase child hunger while giving handouts to billionaires like Elon Musk.

“What is both surprising and alarming is how many research-supporting organizations and institutions outside of the federal government have immediately capitulated to these bans – despite the fact that it’s not even clear whether these orders are legal.“ scholarlykitchen.sspnet.org/2025/02/19/d...

One of the most important facts in deep brain stimulation that we don't acknowledge - the therapy has *always been* closed-loop. medium.com/neuroenginee...

Yep, project manager for #NSF Small Business Innovation Research grants was let go. These grants are the gap between founding and #venturecapital so vital to #innovation and US competitiveness

I love writing. I struggle with it, at multiple levels. But I still love writing.