Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!nntp.sei.cmu.edu!news.psc.edu!hudson.lm.com!newsfeed.pitt.edu!gatech!howland.reston.ans.net!ix.netcom.com!netcom.com!shankar
From: shankar@netcom.com (Shankar Ramakrishnan)
Subject: Re: Unphilosophical investigations
Message-ID: <shankarDBq06M.BJz@netcom.com>
Reply-To: shankar@vlibs.com
Organization: VLSI Libraries Incorporated
References: <804409630snz@chatham.demon.co.uk> <3u50gf$il0$1@mhadf.production.compuserve.com> <805709662snz@chatham.demon.co.uk>
Date: Fri, 14 Jul 1995 19:15:58 GMT
Lines: 62
Sender: vlsi_lib@netcom17.netcom.com

In article <805709662snz@chatham.demon.co.uk> ohgs@chatham.demon.co.uk writes:
>A further paper which may be of interest is JJ Hopfield (Nature  376 33-36)
>I quote the abstract directly:
>
>Pattern recognition computation using action potential timing for stimulus 
>representation. 
>---------------
> A computational model is described in which the sizes of the variables are 
> represented by the explicit times at which action potentials occur, rather 
> than the usual 'firing rate' of the neurons. The comparison of patterns over 
> sets of analogue variables is done by a network using different delays for 
> different information paths. This more of computation explains how one scheme 
> of neuroarchitecture can be used for different sesnory modalities and 
> seemingly simple computations. The oscillation and anatomy of the mammalian 
> olfactory systems have a simple interpretation in terms of this 
> representation, and relate to processing in the auditory system. Single 
> electrode recording would not detect such neural computing. Recognition 
> 'units' in this style respond more like radical basis function units 
> than elementary sigmoid units.
>
>The concept is that there exists a multidimensional space in which the 
>components of a sensory stimulus is located; and that the relative timing of 
>the arrival of signals at key integrating cells defines the locus of events 
>within this space. A recognition cell is triggered by the simultaneous arrival 
>of all of the signals from this space. Thus smell-related data cause excitation 
>within the neurons which support this abstract space and stimuli fly hither an 
>yon. Their relative timing encodes information, however, and the cell which 
>corresponds to 'lemon' fires when the signals that happen to impinge on it 
>coincide in time. 'Lemonness' is both encoded and discovered by this 
>means. Hopfield shows that if the log() of the strength of the signal is 
>encoded in the timing of it, then the decoding is highly versatile and rapid: 
>one architecture fits all, so to speak. Binding -the linking together of the 
>components of a percept -appears to occur through resonant oscilliatory 
>excitation in the respective bits f tissue; and this entrainment might provide 
>the clock against which such time-critical computation might be set.

Interesting. Does this correspond to saying that the qualia or sensation
of lemonic smell is bound to the firing of a 'lemon' cell? 

The case with vision is also quite intriguing. We have several regions
to process vision. The occupital lobes at the back of the head do low
level processing and stimulating these cells typically produce stars
and random flickers in the visual field. However, there are other regions
that are at a higher hierarchical level, and stimulating a cell in these
regions can cause a picture of one's pet or a loved one to appear. Now
it is quite plausible to assume that the lower region does some sort of
processing that causes the corresponding cells in the higher levels to
respond appropriately in response to stimuli from the eyes. The question
is, does the direct stimulation of a cell in the upper region (that may give
rise to the picture of a dog) percolate to the lower levels to form
some kind of a bitmap, or in othert words, is the association two-way?

If the answer is no, then the whole thing is certainly bizzare, for we
have just a single cell (or a small set of cells) whose firing determines
whether we perceive a dog or not, without the bitmap being re-created.
Maybe Penrose is partly right : there may be a deeper structure that
somehow interlinks qualia to activities of individual neurons, so that
the algorithmic complexity of an individual neuron is not of the
order of 1, but virtually infinite.


Shankar
