possible new way to model neural nets?

[Hawley K Rising]

Kevin N Gurney (hssrkng at brunel.ac.uk) wrote:

[lines...]

: well spotted...!

: It is known that there are similarities between certain chemical
: 'diffusion' equations and the formation of topographic maps in
: self-organising nets. This was first spotted by Jack Cowan (I think) in
: the early 70s (date?). Alan Turing (1950s) had played with the chemical
: equations as a basis for morphogenesis in biological systems. People
: like H. Haken are into this kind of thing (his so-called "synergetics") - see
: for example Int. J. of Bifurcation and Chaos vol 4, 1069-1083 1994.
: Apart from the maths, look at the front cover of this issue (chemical
: 'waves') and compare with occular-dominance maps in visual cortex.
: Swindale (1982?) makes a similar appeal to the similarity between zebra
: stripes and occular dominance columns.

The Cowan article I have which looks at them (H.R. Wilson & J.D. Cowan 
1973) does so in context of visual processing, where they appear during 
the collapse of another wave form.  In my own examination of this it 
occurred to me that if the same wave form were applied shifted at this 
point, it would be additively reinforced in the direction of the shift 
and negatively reinforced in the direction opposite, making it a cheap 
motion detector.  I never got much further with the idea, suffice it to 
say diffusive wave propagation could serve a lot of other purposes, 
locally.  It would be somewhat slow as a communication means.

Hawley Rising
rising at crl.com