Connectionists: Neural modeling article

[Douglas S. Greer]

The following article, which describes a new chemical model of neural 
computation where neurotransmitters store information like molecules of 
ink in a photograph, is now available at 
http://www.gmanif.com/pubs/TCS_ISANTF.pdf

Greer, D.S.
Images as Symbols: An Associative Neurotransmitter-Field Model of the 
Brodmann Areas
Transactions on Computational Science V, pp. 38–68, 2009.

The Java code used for the digital simulations of the computational 
manifold automata is available free of charge to academic and nonprofit 
research institutes (See http://www.gmanif.com).

ABSTRACT
The ability to associate images is the basis for learning relationships 
involving vision, hearing, tactile sensation, and kinetic motion. A new 
architecture is described that has only local, recurrent connections, 
but can directly form global image associations. This architecture has 
many similarities to the structure of the cerebral cortex, including the 
division into Brodmann areas, the distinct internal and external lamina, 
and the pattern of neuron interconnection. The images are represented as 
neurotransmitter fields, which differ from neural fields in the 
underlying principle that the state variables are not the neuron action 
potentials, but the chemical concentration of neurotransmitters in the 
extracellular space. The neurotransmitter cloud hypothesis, which 
asserts that functions of space, time and frequency, are encoded by the 
density of identifiable molecules, allows the abstract mathematical 
power of cellular processing to be extended by incorporating a new 
chemical model of computation. This makes it possible for a small number 
of neurons, even a single neuron, to establish an association between 
arbitrary images. A single layer of neurons, in effect, performs the 
computation of a two-layer neural network.
Analogous to the bits in an SR flip-flop, two arbitrary images can hold 
each other in place in an association processor and thereby form a 
short-term image memory. Just as the reciprocal voltage levels in a 
flip-flop can produce a dynamical system with two stable states, 
reciprocal-image pairs can generate stable attractors thereby allowing 
the images to serve as symbols. Spherically symmetric wavelets, 
identical to those found in the receptive fields of the retina, enable 
efficient image computations. Noise reduction in the continuous wavelet 
transform representations is possible using an orthogonal projection 
based on the reproducing kernel. Experimental results demonstrating 
stable reciprocal-image attractors are presented.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: dsgreer.vcf
Type: text/x-vcard
Size: 200 bytes
Desc: not available
Url : https://mailman.srv.cs.cmu.edu/mailman/private/connectionists/attachments/20090810/791f55b9/dsgreer.vcf