Does anyone have anything earlier that they consider part of this framework? Maybe from motor cortex? Something from asymmetric Hopfield networks or Walter Freemen? Even Hebb's cell assemblies were dynamic, but I don't know if they were ever considered in terms of neural state space "trajectories". - ThreadSky

dlevenstein.bsky.social • 5 days ago

Does anyone have anything earlier that they consider part of this framework? Maybe from motor cortex? Something from asymmetric Hopfield networks or Walter Freemen? Even Hebb's cell assemblies were dynamic, but I don't know if they were ever considered in terms of neural state space "trajectories".

Comments

wiringthebrain.bsky.social•5 days ago

Walter Freeman’s work

mattperich.bsky.social•5 days ago

That's the earliest one I know of, at least on neural trajectories. There's a paper from ~ the 70s (I think?) that I've seen that used dim reduction to look at coactivation of motoneurons which starts to look slightly trajectory-like but not fully the same. Can't recall the authors though...

timothyoleary.bsky.social•4 days ago

Goes back to at least the 1960s: Fitzhugh's planar reduction of the HH system into fast/slow manifolds. Then all the literature that followed in neural dynamics. In parallel, in math bio there was a similar subfield of model reduction and singularity/catastrophe theory on population dynamics models.

dlevenstein.bsky.social•4 days ago

Do you see these as a single research program, eg because of the use of dynamical systems tools/ideas?

IMO the dynamics-as-computation program is distinct from (though ofc inspired by) this line of neurodynamics work, due to its focus on computation rather than biophysics.
https://bsky.app/profile/dlevenstein.bsky.social/post/3lirhksshqc24

dlevenstein.bsky.social•4 days ago

Happy to get some pushback here, there’s an argument to be made that once you bring in DST, all neurodynamics is in the same big tent. But it seems like the 2010s dynamics-as-computation movement, popularized by eg mark churchlands work, was a qualitative move enabled by new tech and analysis.

timothyoleary.bsky.social•4 days ago

Here's my pushback: let's say you're studying a neural system that controls movement, precisely modulating timing and vigor of muscles in a state dependent way, a system that organises dynamics into several distinct patterns according to behavioural state, with a decision mechanism for switching

timothyoleary.bsky.social•4 days ago

between them. At what point is this "biophysics" as opposed to "computation"? Does it depend arbitrarily on the number of "units"? The species? I put it to you that there is little or no deep scientific insight or principle that comes out of so called computation-through-dynamics that isn't already

timothyoleary.bsky.social•4 days ago

thoroughly analysed and understood in a "simple" neural circuit consisting of a handful of neurons (sometimes only one neuron!) Computation-through-dynamics seems to be mostly branding. A way to lump (and diminish/elevate) research. The one technical difference is the use of clever dimensionality

mraginsky.bsky.social•4 days ago

From my vantage point of a control theorist long interested in neuroscience, DST is broad enough to account for all the architectural layers, from neurodynamics (eg Freeman, Grossberg, Amit) all the way up to cognition. But this needs care.

mraginsky.bsky.social•4 days ago

One has to be able to cast dynamics as computation and computation as dynamics, for example. There’s a role for control as well, eg feedback, reentry etc. on lower layers and more symbol-like control on higher layers.

mraginsky.bsky.social•4 days ago

Here’s an interesting paper by Roger Brockett that makes some of these notions precise: https://hrl.harvard.edu/publications/brockett94dynamical.pdf

rdgao.bsky.social•5 days ago

XJ Wangs early (~2002) decision making stuff? Not trajectory explicitly (I think?) but fixed points in state space.

And definitely Walter Freeman III

Comments

Posting Rules

Reply