Profile avatar
neuroai.bsky.social
Principal Researcher @ Microsoft Research. Cognitive computational neuroscience & AI. Writer. Nature wanderer. www.momen-nejad.org
977 posts 12,454 followers 487 following
Getting Started
Active Commenter
comment in response to post
I’m really enjoying it. First time in years we have something that resembles the golden sci twitter era. So happy to see colleagues post papers and discuss ideas, even when we disagree.
comment in response to post
We still have distracting nonsense, it’s just more dystopian, no longer fun.
comment in response to post
Oh the paper linked in this other reply addresses some aspects of this: bsky.app/profile/hars...
comment in response to post
Hmm I thought there might also be medial temporal lobe deficits involved (for summoning imagery in general), &/or other sensory regions depending on the type of aphantasia, since it’s not all just visual deficits? @felipedebrigard.bsky.social knows a lot about varieties of aphantasia.
comment in response to post
FWIW, not every single SCB talks that way. @djabaudon.bsky.social and I discussed what the differences (not failures) of organoids can teach us about what cells are capable of if given the opportunity (if in vivo constraints are removed) in this article many years back: doi.org/10.1242/dev....
comment in response to post
Sounds closer to affordances.
comment in response to post
Go Ari!
comment in response to post
I’m taking it seriously, hence the question. In NYC “the school for poetic computation” runs like you say, tuitions fund it, but they also find doners to make tuition free for as many as they can. so a real precedence exists. IMO a dream university could be funded by other sources to make it free.
comment in response to post
Why isn’t the tuition free?
comment in response to post
Beautiful work
comment in response to post
Very likely, yes :) Happy to hop on a call and chat about it. I’ve been especially interested in comparing the mathematics of human exploration and naturalistic settings with exploration in single cell organisms.
comment in response to post
Very cool! Will the data be open?
comment in response to post
Sorry, are you suggesting nuclear energy plus the current miserable state of AI is a good model of biological intelligence? Hard to understand given the word limit in these posts so maybe that’s not what you’re saying, but that statement is simply wrong on so many levels.
comment in response to post
I'm afraid I'm going to get pedantic, but I think y'all are talking past each other, & want to try to clarify, so please bear w me while I state some obvious (or wrong?) things. Shannon info I is only defined over a stationary ensemble, then for a single element x of the ensemble I(x)=-log p(x), &
comment in response to post
Totally Agreed on the 2nd part. To answer the 1st: They use the term read out a couple of times in that paper. But it might have been a colloquial use as the work doesn’t meet the causal criteria you mentioned. That’s the context of my comment.
comment in response to post
Beautiful! Both the image and the finding.
comment in response to post
And the driver of this reorganization? Hippocampal neurogenesis! Eliminating hippocampal neurogenesis prevented engram rewiring and ‘froze’ memories in their original precise state. Conversely, promoting hippocampal neurogenesis accelerated hippocampal rewiring and the emergence of gist. 7/n
comment in response to post
Very helpful, you raise the bar for read-out: 1. Mutual information/MI isn’t proof of read-out. 2. Communication subspace analyses test noise not signal flow. 3. Read-out needs causal evidence/perturbation, phase lag not enough for directionality. So: Shenoy 2013 doesn’t do 3/can’t justify read-out?
comment in response to post
That’s what he’s saying no? I read the comment as proposing causal perturbation is necessary and communication subspace or mutual information are not evidence for readout.
comment in response to post
Very helpful, you raise the bar for read-out: 1. Mutual information/MI isn’t proof of read-out. 2. Communication subspace analyses test noise not signal flow. 3. Read-out needs causal evidence/perturbation, phase lag not enough for directionality. So: Shenoy 2013 doesn’t do 3/can’t justify read-out?
comment in response to post
I think you’re right. Long history of theoretical work in complexity theory, physics, information theory, neuro, psych, philosophy grounds these differences. IMO It takes interdisciplinary scholars to see both POVs, more likely in neuro institutions w intellectual breadth but status quo is ML-heavy.
comment in response to post
No, IMO it's neither semantics nor measurement difference. It's different theories of how the brain works. The more true believers of "read-out or bust" insist there's no difference or mock even the possibility of other theories here, the more likely it is a meaningful theoretical difference.
comment in response to post
Personally I think this discussion is a lot more nuanced than early definitions. It concerns whether there are different paradigms in mathematical accounts of the brain, their differences, lineage, consequences, etc. IMO you shouldn’t worry about your class definitions at this stage.
comment in response to post
I totally agree, computation is in common. I genuinely don’t think computation in itself is the culprit, but how we think the brain is doing its thing. I liked your point that maybe they’re just against digital metaphors. But I’d like to think more to name these views better, perhaps with your help!
comment in response to post
I understand that that’s your perception, and my personal view is closer to yours, but I genuinely can see the other perception as well. As in, I can see and acknowledge it as a different view, not as a simple confusion or semantic disagreement about concepts.
comment in response to post
Sorry you had that experience, but I respect that you still read the papers. We had Earl on the learning salon & he was fantastic even when we disagreed. I personally have never seen him be aggressive at any conference or occasion but I’m not as senior as you are so the dynamics might be different.
comment in response to post
That’s fair. I might change my view about them as well. But right now I feel like I understand what they’re saying, even though I don’t agree with all of it, I respect it, and it feels like a different paradigm.
comment in response to post
I agree we need better naming. They use computation & we use dynamical systems & attractor basins. They may mean the historical ways in which us comp neuro folks model/explain brains. But There is a paradigm difference, maybe you’re an extreme proponent of our paradigm & will never see theirs:)
comment in response to post
Yes, again, we agree on that, But not everybody does! I genuinely believe their viewpoint is valid, Even if I don’t agree with it. That being said there is a very big difference between the analog & digital synthesizer :) & one is not a good model of the other even when they sound the same.
comment in response to post
Well I don’t know the context of that philosopher but that would be a bizarre & inaccurate statement from Yoshua if it was directed at Earl! Physics is actually a lot closer to the dynamical systems approach Earl is advocating for! If anything Earl can make that comment to us lol
comment in response to post
And that can be your valid conclusion within your paradigm! It doesn’t mean that they are wrong. I had a similar trajectory & our frameworks are close, but after more complexity science & doing a bit more math, it has recently become like a Necker cube: I see their view too. We’re not kings:)
comment in response to post
They have a different paradigm than us, cf the analog versus digital synthesizer analogy, but you’re very resistant to even seeing it :) which is normal behavior of someone deep in one paradigm:) Also though IMO mysticism is extremely valuable, it’s not science. Earl’s view is science. Not magic.
comment in response to post
It’s what those of us in the computational view have to do /have been doing. Clearly as someone in the computational framework I find that productive, but not everyone has to agree with us. These are 2 paradigmatic ways of seeing the same phenomena hence the analogy to a Necker cube: I can see both.
comment in response to post
And of course you can still find it compatible! The point is: for primarily computational people like me it is compatible. But not for everyone & important to understand them. My fav philosophy prof used to say we should understand a point so well we could defended in court before criticizing it :)
comment in response to post
I didn’t see what you inferred in what was written and still can’t, but thanks for engaging. And congrats on getting good grades on your recent test!
comment in response to post
It was interesting though to see how hard it is for those of us trained in a predominantly computational approach to see how the other folks don’t find our way of thinking necessary. The convictions, one-liners, & certainty of some comments give a paradigm difference feel. Which is great for neuro.
comment in response to post
IMO no one's talking past you: we read your simple points which echoed Blake, I explained what you didn't understand (e.g. tremors) & noted *we* call this readout but it's not necessary for them. The loop: repeating this cycle. Interesting though, makes me think Earl might have a paradigmatic point.
comment in response to post
IMO best stop the homunculus/other straw-man points, they degrade Earl's deeper point. Not sure how you infer the 2nd point & I don't get it but Earl & some other of us clarified, maybe more careful reading helps? With love, I'd like to exit the loop of one-liners/repetitions. ps. I like your posts!
comment in response to post
I don't think we're making progress, but I'm glad you understand now that they can explain discrete actions without tremors. It's a bit like learning a new language or seeing the rabbit-or-duck. I suggest taking time to immerse in their paradigm to note the point below. bsky.app/profile/neur...
comment in response to post
We're in a loop. Again, those of us in a computational framework find read-out compatible, but for some from the dynamical approach it's not necessary, or it can be seen as misleading. Like you couldn't see the difference between fixed points & a limit cycle, they may not see our paradigm's reading.
comment in response to post
Agreed but I'm mostly in the computational framework like @tyrellturing.bsky.social. IMO @earlkmiller.bsky.social has a valid point: in his view read out is not necessary (or it may be misleading metaphor). I wanna understand those who see this as a paradigmatic difference in understanding brains.