john <John at nine7.demon.co.uk> wrote :
> It is doubely important that people understand them as besides becoming
> important, the human mind is also becoming important as muscle gives way
> to machines and the human mind is neural networks.
quite. thus:
> nn's are analog computer that can have digital inputs and outputs in the
> electronic versions.
artificial neural nets may be (and usually are) represented completely
digitally, including the activation values of the hidden units and all the
weights. the digital values are intended to approximate real-valued
parameters in biological neural nets.
> The easiest way to understand how a neuron works in a network, is to
> compare the neuron with an individual in their social connections.
the most accurate way of understanding how a neuron works in a network is
to examine the neuron's activation and learning functions.
> Like lightening chooses the easiest route down a tree to the ground, so
to
> does inputs find their way to the output action.
what a strange comparison.
> The big problem in electronic nn's is finding the quikest way to choose
the
> conections from the cheapest ways of wired up conections to choose from.
not really. the biggest problem in artificial neural net dedicated
integrated circuit design is scaling. however, there are many other
solutions. many of these are prohibitively expensive or suffer from their
own scaling problems. alternatives to using wires for the connections could
involve radio frequencies, lasers, biological materials, or many other
signal media.
> The method you use to select the right input to prioritise in electronic
> nn's is by the use of algorythms. ...
> In an animals brain the neuron can't geuss whats important, it has to be
> told whats important.
> What tells it I wonder, is geneticly told to keep randomly varying
towards
> a no action default or is there another network that works on the
problem?
i don't know that "guessing" and "being told" are necessarily relevant
here. the neuron's behaviour is presumed (according to scientific thought)
to be characterisable by a finite set of rules, as is the behaviour of
anything. in neural nets, the behaviour of individual neurons contributes
in a highly complex way to the overall behaviour of the net. you might
describe what it is doing as "guessing", or "being told by the algorithm".
i'm not sure if this is your point, but you may only be considering
supervised learning, which is a subset of all types of learning in neural
nets. in this case you might rightly wonder where the error measure is
determined. in some nets it is possible to determine an error measure via
signal feed from another net, which is certainly a prime candidate for some
systems in biological nets. in other supervised nets, the training signal
may derive from an external transducer or from the results of a
self-organising net.
> I wonder if the network of glyle cells is this selecting deciding
network.
glial cells act as insulation for neurons, providing a medium to support
them in and handling their metabolic needs. although they have several
interesting properties and their role in the information-processing system
as a whole has not been completely determined, very few people believe they
are actually involved in signal processing of their own.
dog
|dog at dog.net.uk|http://dog.net.uk| cheeky as buggery