[Fwd: Re: [Comp-neuro] Noise, redundancy and biophysics]

comp-neuro moderator reinoud at castafiore.cde.ua.ac.be
Thu Jul 24 10:59:00 CEST 2008

---------------------------- Original Message ----------------------------
Subject: Re: [Comp-neuro] Noise, redundancy and biophysics
From:    "jim bower" <bower at uthscsa.edu>
Date:    Wed, July 23, 2008 7:33 pm
To:      bard at math.pitt.edu
Cc:      "Sacha Nelson" <nelson at brandeis.edu>
         comp-neuro at neuroinf.org

Great. This is then close, unless I am missing something, to the kind of
analysis that Shannon would have attempted.

Or am I missing something.

Of course the problem is, once again, that "noisy" inputs (the sum of
individual "noisy neurons") suffer from the same fundamental problem that
we don't understand the code -- in fact let me move one step closer to the
cliff with several other questions about the significance of action

Who says that peak responses (as in tuning curves or PST hisograms or the
like) represent the most significant responsiveness in a neuron. Or the
stimUlous that generates the peak the most important stimlous - or the
information that is being transmitted by the neuron to the next neurons?
In biology its the steepest slope of a function where most of the action
likely is.

Second question, in signals represented across multiple channels, the
significance of an event (action potential) is only in the context of the
other channels. And as I suggested before the absence of an event might be
as or more important than its presence. But absence at one moment might be
more significant than at other moments. Now you have some waiting of
significance for different things that didn't happen.

Finally, theorists need to keep three things always in mind:

1) Experimentalists generally seek experimental stimuli and experimental
conditions that produce the most obvious responses (biggest) responses in

2) Experimentalists are generally story tellers and to this day seldom
report the statistics around their results.

3) Experimentalists almost always show the best examples from their data,
declaring those examples to be "representative"


4) Most importantly, experimentalists seldom report all their results and
in particular usually figure out ways to explain away contradictory
results so as not to produce confusion. And the explaining away is done
prepuublication so it never sees the light of day.

This is a corellary to the "tyranny of ideas". It is the "tyranny of a
clear story" and imposed by the story telling nature of most biology.

Not to discourage anyone, this is what computational neuroscience has to
change (and slowly is actually in my opinion).

------Original Message------
From: G. Bard Ermentrout
To: James Bower
Cc: Sacha Nelson
Cc: comp-neuro at neuroinf.org
ReplyTo: bard at math.pitt.edu
Sent: Jul 23, 2008 11:25 AM
Subject: Re: [Comp-neuro] Noise, redundancy and biophysics

I want to once again emphasize that the "noise" that we discuss in our
paper is not meant to be a distraction but rather is the signal in the
sense that we do not know precisely the nature of the summed EPSPs and
IPSPs that are felt at the site of AP generation. Thus, we are using a
broadband input to study how a neuron responds (in some sense, like the
ideas of many others, notably Bialek, Berger, et al) reliably when it is,
in absence of input, firing an a fairly regular fashion. This is a
somewhat restictive environment (admittedly)  and we mean by noise the
form of the signal - as it is repeatable, it is not noise in the
engineering sense of something to be avoided. I agree with Sacha (at least
at the hillock), that channel noise is not an issue. We were interested in
whether a broadband signal could get through reliably in the presence of
other signals that are of different origin (and treated as "noise" with
respect to the first signal).


Sent via BlackBerry by AT&T

More information about the Comp-neuro mailing list