Mark Musolino (990701.1442)
Rick Marken (990701.0915)]
I think this little episode can help us see why real learning
is conceived of (in PCT) as a biased random walk process
(the "E. coli" approach).
Would someone be kind enough to give a brief explanation of the "E.coli
approach". This is the 2nd or 3rd reference to E.coli I've seen this week,
and while I'm familiar with E.coli, I don't know what it has to do with
control issues. Feel free to send something directly to me, so as to not
"bother" the entire group (which may have already discussed this point at
length). Thanks.
Peace.
- Mark
Human Movement & Balance Laboratory
Department of Otolaryngology, University of Pittsburgh
Eye & Ear Institute 128, 200 Lothrop St., Pittsburgh, PA 15213
(412) 647-8069
(412) 647-5392 fax
markmuso+@pitt.edu
[From Rick Marken (990701.2300)]
Mark Musolino (990701.1442)--
Would someone be kind enough to give a brief explanation of
the "E.coli approach".
The "E. coli approach" is a control process that works when
the output of a control system is a random variable. It is
named for the navigational process used by the E. coli bacterium.
E. coli can move almost directly to a food source even though
it can't steer. E. coli can move only in a straight line,
propelled by rotating flagella. It changes direction by "tumbling" --
letting it's flagella go limp briefly and then taking off in a new,
randomly selected direction. E. coli gets to the food by spending
more time traveling toward it than away from it. It does this by
continuously computing the difference between sensed food gradient
and an internally specified reference for that gradient; the
differnece is an error signal. A tumble occurs when accumulated
error reaches a threshold level. The accumulated error builds
up slowly when E. coli is moving toward the food (up the gradient)
so that the sensed gradient is close to the reference (the error
signal is small); this increases the time until the next tumble
occurs and keeps E coli moving up the gradient.The accumulated
error builds up quickly when E. coli is moving away from the food;
this decreases the time until the next tumble and leads to E. coli
quickly moving in a new, randomly selected (but not necessarily
better) direction. This E. coli navigation system is a control
process: it forces a perceptual variable (sensed chemical gradient)
to a reference value and keeps it there (most of the time)
protected from disturbance.
You can see how this process works by trying the "Selection
of consequences" demo at my website:
http://home.earthlink.net/~rmarken/demos.html
The "Control" button runs an E. coli model (in two dimensions).
You can compare the behavior of this model to that of a subject
(the "Subject" button) trying to move a dot to a target (by
pressing a bar that randomly changes the direction of the dot)
and to a model that changes direction randomly and at random times
(the "Reinforcement" button, which produces a "drunkard's walk").
Best
Rick
ยทยทยท
---
Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/