[FRIAM] Who's on Friston? Me and my Markov Blanket
David Eric Smith
desmith at santafe.edu
Sun Dec 6 13:53:56 EST 2020
Steve, hi and thank you,
Luckily, I actually was at that talk, so didn’t have to backfill.
I don’t know what I think. I have been aware of Karl’s work through various enthusiasts for it over the years, but there is such a firehose of volume that I wasn’t willing to start, for a “free energy principle”. I agree that it is good he talks about inference and what I take to be causal reasoning in systems with feedbacks.
I guess I would still like to hear the answer to the question Cris Moore asks: what are you adding, Karl, beyond what Judea Pearl was already doing by putting boundary states between interiors and exteriors in Boolean networks, to define criteria of conditional independence? I don’t know that Karl ever gave an answer to that in which I saw a crisp statement of content.
To the extent that I thought I roughly followed the talk, that particular talk seemed to be concerned with what can be said about steady states, and a kind of “holographic” manner in which the dynamics either within or outside the boundary may be encoded in timeseries of states on the boundary (the Markov blanket). In that respect, the idea seems similar to what the Chaos Cabal back at UCSC (Farmer, Packard, Shaw, Crutchfield) did with “geometry from a timeseries”, arguing, for example, that one could reconstruct aspects of spatial structure in a turbulent flowfield from samples of velocity at a single point in space, but over extended time. It does seem that there would need to be some kind of trapping condition: that state information not be able to flow to infinity at finite rate forever, so that eventually any states however remote would get reflected back onto the (finite, by construction) boundary. I have not tried to think carefully about what kinds of information capacity limits should be needed for that to be possible, and it isn’t something I have ever studied from those who may know a lot about those questions. The notion of finiteness and “reflecting back” seems similar to me, to the way total internal reflection operates in Anderson localization. I don’t know what-all has been done to make mappings between information dynamics on Boolean networks, and continuum or peudocontinuum systems such as Anderson-localizing insulators.
I found Karl’s talk a bit frustrating; it did not have the feel of a talk that was mostly concerned with presenting a tool and putting it in the listener’s hand to understand and use. It was again the firehose, with a sort of faux-bashful admission at the beginning that he always tries to put everything he knows into every talk, and will therefore not finish the narrative he starts. Too many strings of notation without explaining how the reader should know what idea it was after or how that was reflected in the notation. Having also committed that laziness in talks, I am in no position to throw stones, but listening to Karl makes me want to be more conscientious the next time I have to present something.
But maybe it’s all okay. It may be that what he is doing establishes a useful kind of holography principle, mapping currents and state fluctuation statistics from a volume (which could perhaps be indefinitely large) back onto the finite boundaries of interiors in that volume. If indeed the whole state space is infinite, but the information dynamics is trapping, then there should be some kind of large-deviation behavior, such as occurs in reliable coding theory, talking about how more remote volumes, carrying ever-less probability to be occupied, will take longer and longer to have their contributions to fluctuations in the overall state reflected back onto any finite boundary in the interior.
Any results of that kind, however, would probably be available only for steady states. Dynamics would offer a variety of cases growing exponentially in the volume, and I don’t see how they could ever be tamed by projection onto a finite interior surface. As far as I could tell, be only discussed the steady-state case in his talk.
Sorry I do not know how to answer anything of substance. It would take a long slog through a lot of reading. Maybe someday…. I wouldn’t want to discourage somebody else from doing it.
Eric
> On Dec 2, 2020, at 2:52 PM, Stephen Guerin <stephen.guerin at simtable.com> wrote:
>
> Eric Smith and Frank,
>
> Eric, I'd be curious what you think of Karl Friston's talk.
> https://www.santafe.edu/events/me-and-my-markov-blanket <https://linkprotect.cudasvc.com/url?a=https%3a%2f%2fwww.santafe.edu%2fevents%2fme-and-my-markov-blanket&c=E,1,bwsucASTBwpZwb6LvSwhVrtxpPXHBxLMGnhoc2tAOMVV7rQgGVzs-abvO309G1WV6mHPzDHQhdwNYwf-EChJkznwwRh7cKrwguPF15FQjPaRtx_Nqls,&typo=1>
>
> Perhaps better that he moved away from calling his work the "Free Energy Principle <https://en.wikipedia.org/wiki/Free_energy_principle>" and now using the term "Bayesian Mechanics"
>
> Frank, I do find the Markov Blanket/Boundary interesting noting that it originated with Judea Pearl. Did you, Peter and Clark work with Markov Boundaries in your Causal Inference?
>
> -S
> _______________________________________________________________________
> Stephen.Guerin at Simtable.com <mailto:stephen.guerin at simtable.com>
> CEO, Simtable http://www.simtable.com <https://linkprotect.cudasvc.com/url?a=http%3a%2f%2fwww.simtable.com%2f&c=E,1,sBZIROTzFxN2OEExgOO6E27VYkkmbOARQDgQgEwgnKlyqQXmkSP3YM0LJqW7h6DsRwIGi_7AxQaOvih15hiCET5r6cGMyb7hh0bstftC-kQ6sf0rIAYz&typo=1>
> 1600 Lena St #D1, Santa Fe, NM 87505
> office: (505)995-0206 mobile: (505)577-5828
> twitter: @simtable
> z <http://zoom.com/j/5055775828>oom.simtable.com <https://linkprotect.cudasvc.com/url?a=http%3a%2f%2foom.simtable.com&c=E,1,WTFxjbsP13UlpnN21O08TSe1E-dAdcYH9i6xxZT6xcLYZE81BMFk35pKktAuhtVZZr5ylXfpAN41HjQPH197EtVZaEGnfRj4x3EWpWOd9_vjYsm4Gg,,&typo=1>
> - .... . -..-. . -. -.. -..-. .. ... -..-. .... . .-. .
> FRIAM Applied Complexity Group listserv
> Zoom Fridays 9:30a-12p Mtn GMT-6 bit.ly/virtualfriam
> un/subscribe https://linkprotect.cudasvc.com/url?a=http%3a%2f%2fredfish.com%2fmailman%2flistinfo%2ffriam_redfish.com&c=E,1,WlNmm-fL7toVM7m0Cc3KF8cPe8ybYyh6V-c0qN0UH95wJ7j0OnrC-RnCLATfgwHSWRNpEuJ9Mh1JhJbCVhPjTfS9kkXW50hUJNLjfLm3c2fD&typo=1
> archives: http://friam.471366.n2.nabble.com/
> FRIAM-COMIC https://linkprotect.cudasvc.com/url?a=http%3a%2f%2ffriam-comic.blogspot.com%2f&c=E,1,t04HZgncaezjsA8Q89fMpFvlwHT85hPOyPK3Ka9KTrUGrDd8mC37vox45lQU_3iNYTRKIJwzU7Ap7_9TY5QlqAOuq_mLL9hakbCDHAF0IL9l&typo=1
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20201206/fa1e6af6/attachment.html>
More information about the Friam
mailing list