[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 40 Hz RIP

Dear DeLiang

> I meant how many experimentalists have looked for oscillations in the
> auditory system (not the literature).

Galambos, R. Makeig S, Talmachoff PJ (1981) A 40 Hz auditory
potential recorded from the human scalp. Proc. Natl. Acad. Sci.
USA 78: 2643-47.

Basar, E.  (1988) EEG - dynamics and evoked potentials in
sensory and cognitive processes in the brain. In Dynamics of
sensory and cognitive processing by the Brain (E Basar ed).

Sheer, DE  (1989) Sensory and cognitive 40 Hz event-related
potentials. In Brain Dynamics (Ed E Basar). Springer: Berlin.

Pantev et al (1991) Human auditory evoked gamma-band magnetic
fields. Proc. Natl. Acad. Sci. 88: 8996-9000.

Pantev et al (1994) Oscillatory Event-Related Brain Dynamics.
New-York: Plenum.

> The vision community is a better place to voice your complaints,

Now hang on just a minute DeLiang. You have made some grand
claims concerning the application of the "oscillatory
framework" to both visual and auditory processing. Also, as you
will admit, apart from the references I have given above, the
experimental literature you have been using to support your
auditory models is *visual*. It is therefore perfectly
legitimate for us to question the nature of that visual
evidence. We have some very good vision people here in
Manchester with whom I am in regular contact.

If I may remind you, on May 20th you wrote:

> Al writes:

 > > Since one table is red, the firing of RED and TABLE must be synchronized.
 > >  Since the other table is blue, BLUE and TABLE must also be synchronized.
> >  Following the argument to completion implies that all four concepts must
> >  be synchronized.  What we need is two separate instantiations of RED, one
 > > bound to TABLE, the other bound to BALL.

> This isn't really a problem for the oscillatory framework. The
> idea is the following: to have an oscillator assembly
> corresponding to
> TABLE oscillate with double frequency so that this assembly
> synchronizes with
> both the BLUE assembly and the RED assembly. At the same time
> the synchronized
> assembly for BLUE TABLE and that for RED TABLE are
> desynchronized.

> The above idea was used in an early paper on oscillatory
> associative memory
> (Wang, Buhmann, and von der Malsburg, 1990, refs. below) to handle the
> overlap problem, essentially the same as the above problem.
> But the oscillator
> model used there proves to be too clumpsy. A recent model by
> Brown and Wang
> (1996) explicitly addressed the problem of "duplex
> perception" in audition
> using the same idea. But the Brown/Wang model is based on
> relaxation
> oscillator networks, which have an elegant theory and
> computational properties
> behind (see below).

It is clear that the in the first scheme you envisaged the
frequency of the oscillators had only an arbitrary relationship
to the objects or features they were binding, i.e. there is
nothing double about the relationship between TABLE and BALL.
In the second case, as I said, Guy advocated a chaotic
oscillator which has only a remote relationship if any (see
previous message, but see also forthcoming message from Martin

It is certainly not the case that I have a problem with  "a
specific point in a specific paper". I have a fundamental
problem with a major  aspect of the architecture of the
"oscillatory framework" which is to do with the arbitraryness
of the relationship between the oscillator activity and the
objects or features they bind. In other words whether it is
itself a signal or whether it is a sign (to use semiotic
jargon). You are right that there is no point repeating
arguments on the list, and I am sure that most list readers
will have become bored by now, but let me try once more to make
clear what my problem is.

We are agreed that the information for binding is in the
commonality of temporal structure of RF inputs. Now, in the
auditory case the simplest way to measure that commonality is
to keep a copy or memory of each "frame" of the auditory
spectrum over a certain time-window, say 1 second, and compute
a cross-correlation matrix over "channels" in the memory. This
scheme involves no oscillations whatsoever, but it does require
some "delay lines" in order to keep a memory. An alternative is
to first transform the input into the fourier domain, i.e.
sample the power spectrum, and then compute the correlation
between the power spectra for each channel. The advantage of
the second scheme is that it does away with the need for "delay
lines" and the overwhelming evidence is that the cortex,
especially visual cortex, does indeed represent its inputs by
sampling its power spectrum. In order to sample the power
spectrum we do require a set of filters which do have a damped
oscillatory impulse response,  but the frequency of the filter
response is directly related to the signal that is driving it,
i.e. it is itself a signal. Any further oscillatory computation
is entirely redundent.

I rest my case.