Re: limitations of Hawkins' top-down predictive model


Curt Welch wrote:
"feedbackdroids" <feedbackdroids@xxxxxxxxx> wrote:

IOW, the real cortex shows a much greater degree of cross-coupling
between areas than indicated, I think, in the Hawkins' model. Think of
each area receiving FB signals from fully *HALF* of all the other
areas, including many at the same "level" of the hierarchy besides from
many above. Does this really sound like a structure where the main
function of FB is to carry predictive info in a top-down fashion, as
compared to something else, ...

It seems to me that Jeff's "predictive" idea of forming invariant
representations can still be valid, but it's just not as strict of a
top-down structure as he likes to talk about. If a column at any level is
using information from elswhere to improve it's prediction, why not use
information from everywhere you can get it? Why wouldn't both lower level
and higher level concepts be useful in improving the prection of "cat"?
Sensing low level patterns that are like "cat fur" would be predictive of
"cat", but seeing a higher level concept like "cat toys", or maybe "a sign
with the word CAT on it", or "vet office" would also be predictive of a
lower level concept of "cat".

If the network has the power to wire up and use connections from anywhere
that shows a correlation, then you would expect connections to show up from
many different levels.



I think you're exactly right, if you don't take H's simple hierarchical
diagram too literally, as it oversimplifies the real situation. I did
some calculations on the famous Felleman+vanEssen visual cortex
hierarchy diagram. They actually show 11 levels going from V1 on the
bottom to ER [enterrhinal] on the top [with ER being the sole
connection to the hippocampus].

Forgetting about ER, which seems to be a special case, the other 10
levels contain 31 visual areas, with each making bidirectional
interconnections on average with 12 other areas. A lot of pathways
actually span more than 1 level, so the "hierarchy" as shown is open to
some debate. However, the point is that each area performs its
operations in the context of reciprocal feedback from approx 12 other
areas above and below. Some feedback signals will be excitatory and
some will be inhibitory, and/or both might be occuring in every
pathway.

What is interesting, as I indicated last time, is that from the little
that is currently known about nature of FB at lower levels, in MT+V4
back to V1+V2, is that it seems to be the means for providing
wide-field spatial information from "higher" centers back to "lower",
which is not present at the lower levels in the absence of this FB.

How do you think this fits into the "predictive" scheme? [I realize
that asking someone who doesn't believe spatial effects are important
may not be opitimal ;-)].

Also, when we talk about viewing a "coherent" image with all of its
bits in place, as when we look at a visual scene, it seems fairly
obvious that the info carried in these 12 couplings from one area to
another is probably the means for accomplishing this coherency.
Excitatory FB produces some sort of reinforcement/positive-correlation,
and inhibitory FB some sort of discrimination to show off differences.
[personally, I think it is likely this latter effect which produces
visual illusions].

.

Relevant Pages

  • Re: Other "tested" techniques for detecting bias
    ... If you know that an algorithm was used then you don't need to ... You don't "know" until you test your prediction against what actually ... This is the restriction that SETI ... ways to conceal their signals so that we could not decipher them. ...
    (talk.origins)
  • Re: Is the Curt net a kind of decision tree?
    ... only prediction it needs to make. ... maximizing function on it's own, and as a result, there was no need for ... output of the amplifier. ... input signals, to a different set of output signals. ...
    (comp.ai.philosophy)
  • Re: how to measure entropy of music?
    ... I think a vocoder does something similar: linear predictive coding ... To take this back to entropy, the energy of the prediction error could ... This is a common model for music signals, ...
    (sci.math)
  • Re: motion sensor signal compression/prediction
    ... I have a bunch of huge motion sensor signals to lossy compress. ... need random access to the samples block based compression schemes are ... Along with this prediction ...
    (comp.dsp)
  • Re: The Heresy of Scientists
    ... >>>hierarchy with respect to chromosome number, ... >>>hierarchy prediction of evolution, thereby falsifying evolutionism ... >To answer Mark Isaak's assertion that there is no nested hierarchy with ... This also contradicts the prediction ...
    (talk.origins)