From Peter Cariani
Bill Powers (940803.0930 MDT) wrote:
The presence of fine structure in a neural signal COULD be looked upon
as the multiplexing of several kinds of information, but it could also
be seen as the presence of high-frequency detail in a signal where the
basic measure is frequency.
Except that the time patterns yield much, much higher quality
representations than average rate measures and, more importantly,
the way that they change corresponds much better to the observed
psychophysics (which is the way to determine which kinds of
information are actually used). In the case of taste
and color, the adequate electrical stimulus appears to be a time
pattern (DiLorenzo & Hecht, "Perceptual consequences of electrical
stimulaton in the gustatory system." Behavioral Neuroscience 107
(1993): 130-138; Young, "Some observations on temporal coding of
color vision: psychophysical results." Vision Res.1977, 17:957-965.)
Interspike intervals are just the reciprocal
of instantaneous frequency;
I don't think that merely changing from a variable
to its reciprocal can
make any fundamental difference in the process.
Both are continuous
analog variables. The main difference is that
while frequency goes
nicely to zero, interspike interval goes shooting up
to infinity, which is sort of inconvenient.
I have some very basic problems with the notion of "instantanous
frequency". First, instantaneous frequency (i.e. representing
every spike arrival time) is a far cry from average discharge rate
(sometimes called "discharge frequency"), which is mean rate of
spikes produced over an integration window, typically assumed to
be tens to hundreds of milliseconds. I don't think the concept of
"frequency" is well defined unless one includes the time window that
is used to measure it (Heisenberg, Gabor, etc). An individual
spike does not have a frequency, two spikes determine an interval,
but this is not an instantaneous event.
Second, it is actually a very nice property of intervals that
they get longer as the events that they encode are spaced further
and further apart. Oliver Selfridge related to me that they played
longer and longer sequences of "bit noise", and repetition periods
of up to 12 seconds could eventually be discerned by listening.
Longer and longer intervals mean that the same kinds of
mechanisms can handle pitch, rhythm, and even longer range temporal
structures. Recurrent connections automatically get you longer and
longer time delays and lord knows, the brain is full of loops upon
loops.
I would guess that if you can see fine structure in a signal,
that information hasn't been used yet. "Using" it means converting
the information into a signal that
represents it, one signal for each degree of freedom.
One of the big advantages of multiplexing elements is that they can
"pass through" signals that are not processed, and transform those
that are. (Imagine trying to get a message through 10 nodes of
a network of telegraph stations where each station has 1000
inputs and sends out but one signal....)
In a multiplexing network there is still one signal for each
degree of freedom, but each spike train carries more than one
signal (a more reliable way of conveying information,
through populations with more loosely specified connections
rather than through single, dedicated, precisely wired channels).
Anyway, all of this is still in the formative stages,
but the potential power of the general approach
strikes me as being pretty obvious (like using
radio networks and frequency domain multiplexing as opposed to
telegraph-like networks of dedicated lines -- one can do it
with the dedicated lines, but the task is much harder).
While it's still unclear whether the brain uses
these kinds of mechanisms, it's interesting to see
how various parts of the intellectual world
react to the bits of evidence that contradict the cherished
assumptions. I'm sure that by now you're all quite familiar
with that aspect of human nature, as frustrating as it can be.
Best,
Peter Cariani