On Sat, 12 Apr 2003 17:50:51 +0000 (UTC), "Didier A. Depireux"
<didier at tango.isr.umd.edu> wrote:
>r norman <rsnorman_ at _comcast.net> wrote:
>> On Thu, 10 Apr 2003 15:44:25 +0000 (UTC), "Didier A. Depireux"
>> <didier at rai.isr.umd.edu> wrote:
>>> Didier's comments are quite sensible, but I disasgree that 'it is
>> really quite straightforward."
>>That was a bit tongue-in-cheek. The original poster said he would like
>someone to tell him it _is_ straightforward... I wanted to oblige.
>>> Even putting aside the conceptual
>> difficulties of how to do it, the experimental difficulties of finding
>> a preparation on which to test any hypothesis and then actually carry
>> out the work is currently out of the question.
>>I wouldn't quite take it that far. When I characterize a cell in auditory
>cortex, and I hypothesize that it encodes spectral envelope linearly, then
>0) I characterize the cell's properties with a set of sounds.
>1) I verify that I characterized the cell properly by presenting new sounds
>and predicting fairly accurately what the response will be to these
>non-trivial, speech-like sounds.
>So in a sense I verified a hypothesis with a live, awake animal. In
>another, I have not shown that the cell's job _is_ to extract or encode
>spectral envelope, I have only shown that I can correlate very well the
>firing of the cell with some properties of the sounds I am interested in.
>>> Certainly there are specific cases where timing of nerve activity,
>> even to the sub-millisecond level, is critical.
>>That is true, and typically it's periferal. For instance, fibers of the
>auditory nerve can code pure tones' phase up to about 2 kHz (more or less
>depending on who you read). This implies an accuracy in firing time in the
>100 microsec range. By the time you get to cortex, though (which I thought
>was the original question), things are a lot more sluggish and noisy.
>Present the same stimulus 20 times and you will never get 20 times the same
>response, in most parts of cortex.
Of course, you are right. There certainly are specific locations where
specific hypotheses can be tested. I was thinking of a more general
situation, sometimes proposed by people trained purely in engineering,
that neurons can transmit binary data, as in 0-1-0-0-1-0-1 means "37"
where 1 = fire and 0 = don't fire based on some presumed clock cycle.
I particularly like your observation that replications almost never
produce the same results. That is the thing that drives my physicist
colleagues absolutely bonkers. They really don't get why biology
is"so hard." They insist it is because we don't know how to do the
Yes, latencies from a peripheral stimulus get sloppy passing through
intervening synapses. But can't accurate timing still occur even at
the cortical level in the sense of coincidence detection -- as in