remove_this!helbrecht at gmx.net ("Wolfram") wrote in message news:<9n4roo$57kio$6 at ID-40201.news.dfncis.de>...
> Hi Matt,
>> Thanks for the support making clear what in the original posting
> was my goal.
>jonesmat at physiology.wisc.edu (Matt Jones) wrote:
>> Your following paragraph is on that level i was looking for. Now i
> can hook and go more to the details.
>> > The task of a neuron is to move a signal from point A to point B,
> > while possibly performing some sort of operation or transformation on
> > the signal as it goes through.
>> Is that precise? Does it mean, that one specific signal coming
> from A (= input signal?) going to B (= output signal?) is anyway
> identifyable on its way from A to B? - Yes, you wrote, possibly
> the neuron performs some sort of operation ... on it. Does that
> mean, therefore, it is not identifyable on its way?
Well, it depends on what you mean by "identifyable", and also it
depends on what tools are available to do the identifying.
Consider a neuron with a very small number of inputs, say, 3 (as
compared to the true number which would be thousands). If you want to
identify what output corresponds to what input, then a few conditions
must be met:
1) There must be -something- different about each input signal. They
must differ in size, shape, sign, timing or location on the dendrite.
2) The neuron must -somehow- preserve these differences as it passes
them down its dendrites, through the soma and on to the output at the
end of its axon. This doesn't mean that the size, shape etc, must be
exactly preserved. Generally, they are not. But something unique to
each input must be manifest at the output.
3) We must have some way of figuring out -what- it is at the output
that marks each incoming signal.
If these conditions are met, then yes, the signals would remain
identifiable. However, it is currently an actively studied (and
heatedly debated) question whether these conditions are met, and
whether there is a way for us to extract the information about each
input from the output. These studies generally go by the name "Neural
Coding", and in plain language, seek to understand -how- a neuron
encodes information about its inputs into its output, -how- another
neuron (or an observer) might decode that information, and -what- the
meaning of that information is for the context of the neuron, the
circuit that it lives in, and ultimately the behavior of the animal.
In information theory terms, this question would be considered a
problem of a "multiuser channel", or a "multiple access channel", in
which a number of senders (i.e., presynaptic neurons or sensory cells)
wish to send information over a channel that they all share (i.e., the
neuron), so that it can be properly understood at the other end (i.e.,
the next neuron or neurons) with a minimum of distortion or
interference (and no, i don't mean quantum wavelet holographic
oppositional oscillon interference, I just mean noise) by the signals
sent by other senders. This problem has not yet been solved for
telecommunications systems, and certainly not for the nervous system.
So theoretically, it -might- be possible, and true, to say that
signals A and B remain identifiable, in some sense. But it may also be
the case that the brain doesn't really care about the individual
signals very much, and just wants to compute some sort of an average
over all signals, and make an output based on that.
Practically, there are also a number of issues. First, recording
neural signals from many neurons simultaneously appears to be
necessary for answering this question. It is possible, though not
easy, to do this. However, in multielectrode recordings, it is -not-
usually possible to know for sure which of the many neurons being
recorded are connected to which others. So before answering the
question about whether the inputs are identifiable at the output, it
is first necessary to -identify the inputs themselves-. This is
possible in some cases, particularly when only two or three cells are
involved. But recording from only two or three cells doesn't give a
very good picture of what sorts of noise and interference would be
going on in the intact system, where everybody is talking at the same
time. It's a little like eavesdropping on a conversation between two
people, and then trying to decide what is the national policy of the
country in which they live. Not very likely to yield accurate results,
unless some special tricks can be employed to help in the
Apparently, I can never talk about neural coding without putting in a
plug for my favorite book on the subject, so here it is:
Spikes: Exploring the neural code. By Rieke et al, MIT press 1997
Chapters 1 and 2 give a great introduction to the main issues, and the
rest of the book's not to bad either.
Also, there's a much older literature on the issues of parallel and
distributed processing in the nervous system (actually, it's mostly
about artificial neural networks, which -are not- the same thing as a
biological nervous system for many reasons). Thius book is a probably
the original reference, but I'm sure others here can suggest more
recent, better or gentler introductions:
Rumelhart, J. L. McClelland, and the PDP Research Group (Eds.).
Parallel distributed processing: Explorations in the microstructure of
cognition. Vol. 1: Foundations. Cambridge, MA: MIT Press.