> My understanding is that these cross-correlograms have mostly been
> used to uncover underlying connections between neurons rather than
> for examining stimulus-dependent response patterns. This is the
> reason that the shuffling is done; to weed out those nasty
> stimulus-driven synchronies. But what if the stimulus-driven
> synchronies are what the system uses to encode edges and spatial
> intervals between edges? One has thrown out the most important
> aspects of the neural response.
A very important distinction to keep in mind is the difference between
stimulus-TIMELOCKED and stimulus-DEPENDENT. What the shuffling does is
get rid of "synchrony" that is timelocked to stimulus onset. But this
does NOT get rid of all stimulus-dependent synchrony.
For example, as Jim Bower pointed out, the synchrony theory of the
binding problem postulates that some stimuli will make certain groups
of neurons synchronize to each other, in a manner not time-locked to
the stimulus onset (the neurons are only time-locked to each
other). Thus the shuffle will not get rid of this synchrony, and it
would be revealed in shuffle-corrected correlograms. But because it is
only certain stimuli which give rise to this kind of synchrony, the
synchrony is certainly stimulus-DEPENDENT. In short, this is an
example where the shuffle is used to reveal stimulus-dependent
synchronous response patterns and is not used merely to reveal
anatomical connectivity. For completeness' sake, let's cite Singer
and Gray, Annual Review of Neuroscience 1995, on the synchrony theory
of binding.
> I have never understood why synchronies that are not related to the
> stimulus are so highly prized, but those related to the stimulus
> are routinely discarded. Could you shed some light on this?
This question about stimulus-locked synchrony is (in my view) an
excellent one. Stimulus-locked synchrony, by virtue of being
timelocked to the onset of the stimulus, can actually be seen in
PSTHs: if two neurons are timelocked to the stimulus onset, you don't
need to compare one neuron to the other, you just need to compare
their firing times to the stimulus onset. A number of investigators
have begun looking at the question of "how fast can PSTHs follow
stimuli, and how deep into the brain can this fast following go?" for
vision. This is probably something that has been well-studied in
audition, you would likely know much more than I about that end of
things (I'm not an audition person). But for vision, two papers about
this are Bair & Koch, Neural Computation (two years ago?), and more
recently, Buracas et al., Neuron, sometime late last year. Both
concern MT, and both show that a rapidly time-varying motion stimulus
can lead to PSTHs in MT that are very rapidly modulated, with PSTH
peaks that can be as narrow as 5-10 ms. Thus rapidly time-varying
motion signals lead to quite highly (stimulus-timelocked) synchronized
firing in MT. The Meister group also has some awesome papers about
this w.r.t. the retina, and I understand that Wyeth Bair and
colleagues in Movshon's group are looking at similar questions in
V1. Perhaps this is just to say that if stimulus-locked synchrony used
to be ignored, it is not being ignored any longer. We live in a
rapidly time-varying world, and the basic question being addressed by
these groups (how much of that rapid time-variation can be represented
in the brain?) is, in my view, very interesting. (If anyone from
these groups is listening, and I have misrepresented your positions,
please do pipe up and correct me.)
best,
Carlos.