In article <3740B614.BCA68119 at banet.net>, ken collins <kenpc at banet.net> wrote:
>and, btw, this sort of system is easily, and vividly, simulated
>via an approach that makes use of red, green, blue (RGB) color
>mixing to encode "ionic conductances" (in a non-analog machine,
>the color representations are still "state"-bound, but, if one
>wants to, the calculation can be extended to any degree of
>accuracy by just using arbitrarily-long numbers to interpolate
>between the 16M color states of a PC, and having a scalable graph
>window to display the interpolation).
>>this system of stateless calculation breaks all, so-called,
>"NP-complete" "barriers"... is limited only by global i/o channel
>capacities... whole databases can be taken in, and their main
>significance, calculated immediately (in linear "time").
I think you're misusing the terminology -- and concept -- of
"NP-completeness" in your rush to make an incorrect point.
First of all, neural networks, whether artificial or natural, are
of course highly parallel systems, which makes it difficult to
import technical analysis tools such as NP-completeness without
substantial modification. "Obviously," 2^N Turing machines will
solve a problem faster than N will, but this doesn't have anything
to do with non-determinism per se.
But furthermore, the claim that NN's are somehow "stateless" and
hence only I/O bound is incorrect; most interesting NN's incorporate
some sort of state/memory. Furthermore, the reduction of NNs to
Turing machines, in both directions, is well-known. Neural networks,
along with RAMs, Game-of-Life, and so forth, are simply yet another
Church-Turing computation device. You can't claim that continuity
provides some sort of mystic power, as NNs aren't continuous
computers; ionic flow is limited, eventually, to the flow of single,
quantized, ions.
-kitten