Date: Tue, 24 Nov 1998 11:11:12 -0800
From: Chris Cherpas <computeruser@HOME.COM>
Subject: Re: Conrol of (and with) rate of firing
[From Chris Cherpas (981124.1111 PT)]
Clem McGowan (981123.1850)--
>Take a look at the general interest article located at:
>
> http://www.salonmagazine.com/21st/feature/
The emphasis on variations in rates of firing sounded like
an exact fit with one of PCT's modeling assumptions. Is
there another interpretation?
[From Peter Cariani (981125)]
Hi everyone!
One should never try to extrapolate what neuroscientists are
really thinking from a popular science article -- the writers
almost invariably mangle whatever the scientists tell them in
order to get something concise that is semi-intelligible
to the general public.
But of course this is very interesting work indeed. Grey Walter
in the 1950's had such feedback experiments going where EEG's
were displayed and stimuli were triggered on particular phases
of EEG's. (see The Living Brain, c. 1959). There are some
very striking effects that occur when external stimuli are
coupled to endogenous rhythms. (There is a cottage industry
of stimulus-induced brain manipulation, see Hutchinson,
MegaBrain, c. 1991 for a pop-intro).
One needs to be careful here. From the article, there is an
audio feedback of the firings of the neuron(s) and there is
also feedback from cursor movements. First, firing rate may
be under adaptive control, but it may be one of any number
of spike train parameters that could be controlled. Other
potential variables might be the latency of spikes relative
to some reference auditory event or the synchrony between
spikes produced within some neural population. Many of the
putative codes in the CNS also have some component that
varies with firing rate, so that the controlled variable
(I hope I have this terminology right) is not necessarily
exactly the variable that is being used by the system to
encode the percept and/or to actuate a muscle response.
For example, an audio signal with more spikes is louder than
one with fewer ones, and rhythmic firings or bursts may be more
perceptually salient than other patterns.
Usually neuroscientists see some variation of a stimulus
parameter and firing rate, and then, since they can now tell
a story that everyone understands, they stop looking for
alternative codes or conditions that would falsify their
simple story. For example, the bit about the whiskers is
much more complicated than is related in the news story --
much of that kind of work was done under heavy anesthesia where
there is much less activity than in the waking animal, where
there is considerable spontaneous activity. In awake animals
when whiskers are briskly stroked, my understanding is that
many regions in somatosensory cortex are activated (way
beyond restricted classical receptive fields comprising
one whisker). I would point to Nicolelis' work (he uses
50-100 chronically implanted electrodes in an awake animal),
but very similar kinds of things are seen in all other
sensory systems -- the areas that are activated can be
quite broad when the stimuli are well above perceptual
thresholds. These effects cast doubt on simple rate
codes, though not more complicated ones.
It's important not to overestimate our present knowledge
of neural codes in the cerebral cortex -- we are really
in a state that is comparable to molecular biology before
the nature of the genetic code was understood
(e.g. that genetic information resides in DNA rather than protein).
Peter Cariani
P.S. This article set off some other bells.....
What is also interesting about this feedback
experiment is that the time patterns of spikes
that are transmitted over the audio channel in effect
recreate similar time patterns in the auditory system
which then may reinforce the original pattern of
produced spikes (it would be quite interesting if
particular patterns emerged over time, having been
reinforced by the feedback loop). The clicks that
are produced when listening to spike activity we
know robustly create similar time patterns over
large parts of auditory cortex and indeed over the
rest of the cortex (acoustic transients and patterns
of transients are particularly well-represented in
the cortex).
(I think of this because I have been working on
recurrent timing models for auditory expectation,
e.g. how a rhythm builds up an expectation of
its continuation. Movements similarly contain
perceptual expectations of what is to happen as
their consequence).
Let's think about the full percept-action loop
(or action-percept loop, if you prefer) for a minute.
Whenever there are motor commands to activate
muscles, there must be a volley of well-timed
signals to the various muscle groups to initiate,
modulate and terminate movements (to an agonist
at time t, to an antagonist after some delay, etc).
There are efferent copies of the time structure
of this command-volley that are generated, then
there is a similar temporal pattern that is
generated by the stretch receptors when the
muscles move, then there is yet another similar
temporal pattern that is generated in the
environmental disturbances that are produced
(e.g. acoustic disturbances created by
articulatory muscles), then there is a
sensory signal (e.g. an auditory signal of
the utterance) that once again carries many
of the gross aspects of the time structure
of the command, efferent-copy, stretch-feedback,
articulator movement, and acoustic production
(albeit with different characteristic delays).
It may be the case that the whole mechanism by
which we learn to coordinate our bodies has
to do with the conguence/incongruence of these
various time patterns -- that the basic adaptive
mechanisms are set up to put all of these patterns
into congruence, and that the audio feedback is
then a particularly efficient way of closing
the sensorimotor (motorsensory) loop. There is
older work by Morrell, John, and others that
time patterns are assimilated during conditioning,
and that memory traces encode the respective
times of events in addition to their simple
conjunctions, so the general idea is perhaps
not so far-fetched.
Perhaps a test of this would be if the spike rate
were converted into a continouous pure tone whose
frequency varied with spike rate --
would the feedback be as effective?