From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!wrdis01!news.cs.indiana.edu!spool.mu.edu!uunet!trwacs!erwin Mon May 25 14:06:24 EDT 1992
Article 5761 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.ecf!utgpu!csd.unb.ca!morgan.ucs.mun.ca!nstn.ns.ca!wrdis01!news.cs.indiana.edu!spool.mu.edu!uunet!trwacs!erwin
>From: erwin@trwacs.fp.trw.com (Harry Erwin)
Newsgroups: comp.ai.philosophy
Subject: Neuroscience Workshop
Message-ID: <592@trwacs.fp.trw.com>
Date: 20 May 92 02:15:11 GMT
Organization: TRW Systems Division, Fairfax VA
Lines: 53

I'm posting this half way through an exhausting and exciting workshop on
behavioral and computational neuroscience at Georgetown University. The
highpoints so far include a pair of papers by Judy Dayhoff and Stewart
Hameroff and one by Bruce MacLennon.

The Dayhoff and Hameroff papers discuss the neuronal cytoskeleton. The
motivation is work by Gelber (1958), Applewhite and Gardner (1973), and
Fukui and Asai (1976) training Paramecia to run a very crude maze. The
indications are that the cytoskeleton (consisting of an extensive network
of microtubules and other elements also has a function as a nervous
system. This system in the neuron extends (unidirectionally) down the axon
and bidirectionally in the remainder of the cell, connecting the cell body
and the dendrites. There is evidence that it can transmit information on
several timescales.

Judy Dayhoff in her paper reviewed the properties of the MT and identified
a mechanism using the them to transmit error information from the synapses
at the end of the axon back to those on the dendrites, thus supporting the
backpropagation model.

Stewart Hameroff went further in his paper, discussing the several
timescales of data propagation along the cytoskeleton. One timescale
involved propagated conformational change, with the MAP2 elements serving
as a switch when phosphorylated. In addition, the cytoskeleton organizes
the dendritic spines via actin molecules. These allow the synapses to
serve possibly as analog switches, activated on several time-scales. Also,
the MAP2 elements may be able to serve as repeaters to facilitate the
transmission of coherent calcium ion waves. 

The MacLennan paper reviewed the properties of biological dendrites and
synapses, and identified a good deal of evidence that the networks of
dendrites, linked by synapses, are effective computational units. The
evidence appears to be that they are significantly more effective than the
standard M-P neuron. 

There seems to be fairly strong evidence from the experimental side that
the following aspects of Karl Pribram's theory are holding up:

1. The neural (axonal) network is regularly and sparsely connected, serving
as a communications medium.
2. The dendritic network is randomly and densely connected, serving as a
computational element, and
3. The axon hillock serves to summarize the results generated by the
dendritic network, and is the primary non-linear component of the system.

One question that emerges is whether real dendritic networks are
specialized to compute Gabor functions.

Your exhausted servant,

-- 
Harry Erwin
Internet: erwin@trwacs.fp.trw.com


