erwin at trwacs.fp.trw.com
Wed Dec 22 08:14:59 EST 1993
This speculative posting attempts to define a possible engineering concept
for the architecture of the neocortex.
I posit that the neocortex is made up of modules ('loops') consisting of
pairs of KII systems arranged to form feedback systems. The general
structure is as follows:
|/ \ |
(Here, capital 'A's represent upward arrowheads.)
Here, a is the afferent signal to the loop; e is the efferent signal; r is
a reafferent signal from the next loop; t is reafference to the preceeding
loop; fx is the signal to Ky from Kx; and fy is the signal to Kx from Ky.
Both the fx and fy paths preserve topology. Kx and Ky are KII systems
(in the sense of Freeman). Kx supports habituation and conditioning,
with the fy signal being paired with the afferent signal for conditioning.
The fy signal supports attention, and is modulated, supporting a
stochastic resonance process. If there is no fy signal paired, the system
habituates. The Kx system operates as a content-addressible memory to
identify patterns (features) in the afferent signal. The efferent signal
identifies those features present in the afferent signal and attended to.
r controls attention. In networks, these loops are naturally chaotic.
Loops are 'stacked' in structures similar to the following:
e_n mixes the efferent signal from Kx_n to produce the afferent signal to
Kx_n+1. r_n necessarily inverts that mixing to transform a pattern
attended to in the n+1th level into its constituent features in the nth
The weakness of a system designed this way is that attention in each layer
is controlled by the layer below it. Eventually, this regression of
control peters out, and you get a hard-coded attention function. There are
social games that cannot be effectively played by such a system. Hence, a
purely loop-based architecture is not a good model for our brains. At some
level, the efferent signal of a loop has to be redirected into the loop as
a portion of its afferent signal, and a portion of the control signal it
generates (r_n-1) has to be redirected into it to serve as its input
control signal (r_n). Bluffing seems to require more than just context-
sensitivity, so--guess what--it looks like there's a Turing machine in
Internet: herwin at cs.gmu.edu or erwin at trwacs.fp.trw.com
Working on Freeman nets....
More information about the Neur-sci