Following up on Nick's question about "how does the computer know how
it's doing?", here's a recent article from Gerald M. Edelman and
company.
http://www.pnas.org/cgi/content/full/102/6/2111Darwin X: the 12 CPU Beowulf cluster takes sensory input from the
mobile robot, simulates 200ms of neural activity in 200ms realtime,
writes a trace of the activity to disk, and tells the robot which way
to turn. The robot is running a simulated water maze, swimming until
it finds the platform so it can climb out and rest. The neural
architecture is schematically presented in Figure 2, the numbers and
parameters of neural units are tablulated in supporting Table 2, the
topography of the interconnections is detailed in supporting Table 3,
and you can watch the mobile unit in action in supporting Movie 1.
The median time to find the platform improves from 500 seconds to 200
seconds over the course of 16 trials, repeated for 9 different
instantiations of Darwin X which differ in the exact details of the
neural wiring while following the general recipe.
How long it took to analyze the data from those 144 trials and decide
"how the computer was doing" isn't stated.
-- rec --
Locked
