Mark Histed
@HistedLab
Followers
2K
Following
5K
Media
387
Statuses
4K
Aim to understand brain networks and brain wiring, using lasers & neuro AI. Lab head, @NIH. Prev: policy for @democracypolicy. Personal views. Not a lab acct.
Washington, DC
Joined May 2013
New preprint, on 'sequence filtering'. Led by @CianaDeveau, Z Zhou. We see this as a key step forw on how cortex works. All cortical areas have dense exc-exc recurrent connectivity. What do these connections do, esp in sensory ctx? Our data say: they do dynamics/time. 1/4
6
38
174
And I’m big fans of Turing, von Neumann, Shannon, Metcalfe, Hinton, Sejnowski, and all the other algorithmic and computation people. No shade. But data limits still limit understanding of the brain. That’s why #NIH US BRAIN Init was so important: pub investment made new tools.
0
0
4
Same for neuroscience. The lack of ability to measure many neurons’ activity, perturb them, and measure intracellular processes and connections is what limits understanding the brain. The key barriers are not algorithms or AI.
Francois usually has good takes. But this suggests a bit of cluelessness about what the key barrier to progress in biology is. It's not algorithms or AI. It's still a lack of the ability to measure many important things in cells ie. assay techdev. Perturb-seq is not all u need.
7
1
48
Our new paper on large scale holographic read-write experiments observing and controlling thousands of neurons in mouse visual cortex reveals a new functional cell-type that detects even moderate levels of excess activity (just 50 extra neurons firing) then inhibits top down
2
25
128
Great quote
This is insane and the retroactive part would defund most physics, EE, CS in the entire US. I said this once to the national security council at a White House meeting: “we beat our competitors by draining their brains.” This strategy has underwritten our success since WW2. We
0
0
1
Both neural networks and “unitary executive theory”/administrative law. 😭
Can't believe my daily job involves thinking about both polynomial functors and autoformalization of legal statutes
1
0
0
One of the things that gets me down walking around NIH these days is how the buildings are crumbling. This is not just a Trump thing: we have had years of underinvestment in maintenance and too little gov’t in-house capacity, leading to contract work with too little oversight.
0
0
4
Yes! And when neurons (or networks) carry history and change their responses based on a combination of history and input, that is itself a form of prediction.
Prediction is vital for the brain, but not everything. When observations are noisy, estimation improves by looking back. We argue that even single neurons can be retrospective—like LGN lagged cells and olfactory bulb mitral cells:
0
1
3
yes. The 'unitary executive theory' is just the 'president-is-a-king theory', and for obvious reasons that runs at odds with the Framers and the Constitution
While nothing the Supreme Court (the genteel wing of Trumpism) does should be surprising, the court's expected adoption of the "unitary executive" theory in full measure will still be brazen because it has basically no support in the constitution or historical practice
1
0
1
exactly correct - it stalls
What happens to discovery when federal research funding stalls? Science can't wait. 🎧Join Salk’s Beyond Lab Walls podcast tomorrow, October 30, at 9:00 a.m. PT/12:00 p.m. ET for a special *live* episode exploring how the government shutdown impacts our research and the best way
0
1
1
🎯
@willyakah @ron_alfa Anyone who thinks ML engineering is the main bottleneck in building better cellular models is going to be sorely disappointed. Better problem formulation & the right data collection strategy will win any day of the year.
0
1
33
@willyakah @ron_alfa It's also becoming amply clear that collecting huge amounts of perturb-seq data from single gene knockouts or arbitrary drug perturbations with no temporal component in arbitrary cell lines & throwing a billion parameter transformer at it, barely moves the needle as well.
1
3
69
@littmath Physics might be harder than math for AI:
The return of the physicists: "CMT-Benchmark: A benchmark for condensed matter theory built by expert researchers." https://t.co/xMtJprbF9S A set of hard physics problems few AIs can solve. Avg performance across 17 models is 11%. Problems range across topics like: Hartree-Fock
0
2
5
I’d be happy to have Bari Weiss loudly insisting on the free speech rights of government employees in this moment. <taps earpiece> oh, she doesn’t actually care about free speech? sigh https://t.co/QAaLYXjLXl
stringinamaze.net
Journalism's biggest fraud takes the reins at CBS
0
1
0
this is from Anthony Lewis, author of Gideon’s Trumpet (on Gideon v Wainwright.) https://t.co/uMlavge5bx
en.wikipedia.org
1
0
0
The people who built the US Constitutional framework cared a huge amount about info environment. “[The people] have a right, an indisputable, unalienable, indefensible right, to knowledge about the character and conduct of their rulers” - John Adams. 1/
1
0
1