neuromorphicboki.bsky.social
Mathematical Neuroscientist. Biologically inclined.
PhD cand, HEx Lab @ Unimelb, Aus
Research Eng, Dreamteam @ ICM, Paris
emergence, complexity, consciousness, large-scale brain dynamics/modelling, causality, AI, abstraction, mechanistic interpretability
71 posts
567 followers
473 following
Regular Contributor
Active Commenter
comment in response to
post
Heresy! LLMs are nothing like biological systems. #watchthisspace
They do come with a separate set of issues indeed if taken too literally, I agree.
comment in response to
post
Agreed. I think the problem is the syntax and semantics. I’m for letting go of mind-first definitions (a la W. James)—attention, consciousness, etc.— for a more brain-first definition of function (a la Buzsaki). I think mapping from this inverse direction might yield nifty results.
comment in response to
post
My boy Dürer, for sure.
comment in response to
post
This is all possible due to the emergent effort of a collective group of organisers that I'm lucky to be constituent part of.
Ruben Herzog
Mar Estarellas
Ivan Mindlin
Samy Castro
And our fantastic Scientific Chairman, Claudio R. Mirasso
comment in response to
post
🚀 Its an Exciting Opportunity! 🚀
Apply with your abstract by March 14, 2025 for one of 35 spots at the Universidad de las Islas Baleares residency (June 22-26)! 400€ covers lodging, meals & an excursion. Selection by end April – don’t miss out! #CallForAbstracts
comment in response to
post
There will be some incredible keynote speakers, including:
Fernando Rosas @frosas.bsky.social
Lauren Ross @laurennross.bsky.social
Antonella Tramacere
Ivana Konvalinka
Carmen Miguell
Leonardo Gollo
David Sanchez
Lucas Lacassa
Borjan Milinkovic 🤭🤗
Pedro Mediano
Julia Mindlin
comment in response to
post
The workshop is intended to bring together varying scientific disciplines, across scales, to discuss the current best-practices and tools used to study emergent phenomena.
We believe this is necessary and fulfils the original motivation for studying emergence: for true, *scale-integration*!
comment in response to
post
I feel category theory will be useful here. Particularly Chu constructions and spaces.
comment in response to
post
Hard for a single post, but I don’t think computation should only imply manipulation of discrete symbols. I feel we have moved beyond this already—but without formalism. I.e. Neural comp., mortal comp., neuromorphic archiectures etc.
But my general intuition is the relationship b/w info and comp.
comment in response to
post
I’ve been sprinting through these lectures too recently. Completely agree. Concept of computation is in need of expansion.
comment in response to
post
I keep repeating this as the one reoccurring topics worth addressing: If there is at least one true ‘biological law’ it is that it seems biology has learned to exploit physics to create niches.
comment in response to
post
Good read! I would also add to this train of thought Kauffman’s work in general. His book, A World Beyond Physics, has left an indelible impression on me.
comment in response to
post
Hell yeah..!
comment in response to
post
One we recently published. 😊
Our method precisely captures orthonormal subspaces that are dynamically closed from the underlying original multivariate system dynamics.
Would be very happy to discuss further. 🤓
comment in response to
post
At the end, brains brain because of their heterogeneity in structural elements. ANNs ANN only on the basis of a single subunit.
comment in response to
post
And the different cell types (what @macshine.bsky.social, and @jaanaru.bsky.social touch on)..
that makes the brain, *brainy*.
I don't disagree, ANNs computational units might be good abstractions of neurons.
*But* this does not entail that the systems generated by them, are analogous too.
comment in response to
post
This is at complete odds with what Ilya proposes. It is *precisely* the multiscale dynamical structure that exists, not in parallel with, but *because of* metabolic costs of neural spiking (what @pessoabrain.bsky.social mentioned)..
comment in response to
post
Actually, it appears that most effective information exchange may not rely solely on spiking neurons. But it is non-spiking neurons, which effectively coarse-grain the signal, that seem to provide a more energy-efficient and information-rich strategy for the brain to perform its functions.
comment in response to
post
And here, where they actually measured it(!):
comment in response to
post
This was literally shown in many papers, of which this one is a beautiful example.
comment in response to
post
Key results, which altogether have been lost on the ANN and computational community is that non-spiking neurons with graded potentials play a key role in *optimally* routing information in the brain by decreasing the metabolic cost and increasing the information transferred.
comment in response to
post
The brain functions the way it does because of *constraints* placed on it by the metabolic costs of neuronal activity. Not by "more"
And its not even about spiking neurons(!) One of the key drivers of what the brain can do under such intense metabolic constraints are neurons that *don't* fire.
comment in response to
post
To infer what Ilya inferred at #NeurIPS, is wrong. And scientifico-culturally dangerous given his level of pull.
In fact, this whole argument--premised on the notion of using more neurons to get to the brain-like function is--to me--wrong.
comment in response to
post
Beyond the *neuronal* doctrine. ;-)