computer-go: speculative introspection

[William Harold Newman <william.newman@xxxxxxxxxxx>]

I was working on some flooding algorithms and thinking gloomily about
how many CPU cycles they're going to consume, and I got to wondering..

How it is that I can step up to a Go board and recognize quickly that
there's a solid white connection between point A and point B? Even
when the connection involves tens of stones, my wetware seems to
notice it in perhaps 500 milliseconds. Does anyone have any hints as
to how a network built of neurons with switching times on the order of
milliseconds could be arranged so that it notices such connectivity
that fast?

I imagine that this is something that serious neural network and
vision processing people have written about, since IIRC one very
influential critique of perceptrons was that they couldn't recognize
connectedness. Does anyone have any literature references for networks
which recognize connectedness efficiently?

(Of course, it's also interesting to think about how to train a neural
network to recognize connectivity, or how to design the genes that
control brain development so that the network is predisposed to
recognize connectivity. But at this point I'd be satisfied just to
understand the end result: what a neural network which recognizes
connectivity efficiently could look like.)

-- 
William Harold Newman <william.newman@xxxxxxxxxxxxxxxxx>
software consultant
PGP key fingerprint 85 CE 1C BA 79 8D 51 8C  B9 25 FB EE E0 C3 E5 7C