KR control

[From John Anderson (961015.0430)]

[From Rick Marken (961014.1100)]

John Anderson (961014.1035 EDT) --

I think that the r3 in your diagram is unnecessary. And your e3 seems to
function in exactly the same way as my r2; it changes the "bias" for
classifying the perception (p2). My r3 was setting a reference for an
"intrinsic" error (the average error in the KR control system) and adjusting
the perceptual function (S2) on the basis of the size of that error.

Well, now that you mention it, thats right, r3 IS unnecessary in my
diagram, but then it would be in yours, too. I think we both include r3
because we want a separate reorganizing system evaluating whether or not
to change S2.

Your model doesn't adjust the perceptual function...

Thats easy to fix:

                                r3 (zero error)
              r1 ("correct") |
              > >
              v v
       p1--->|C1|------e1----->|C3|---------------------------e3
       > >
       > >
       > >
       > >
       > ----------p2---------- |
       > > > >
       > > > >
       > > > >
Subj | v | |
______|S1|_____________________|O|____________________|S2|<----
Env ^ "restriction present" ^
       > "restriction absent" |
       > > >
       > > >
       qtest(controlled) v qairway
    "correct" <---------- |experimenter| (not controlled)
   "incorrect"

Now if theres an error e1 at C3 the perceptual function S2 itself is
changed in some way before or during the next test. In this diagram
Ive also removed IO, since its function can be included in S2. I think
I understand how this system would work. As Bill Powers (961003.1100
MDT) said:

My guess about this experiment would be that the restriction is sensed in
terms of the extra effort required to maintain respiration at the required
depth and rate.

So the subject is sitting there, breathing through this tube, and has to
decide whether the restriction is there or not. If it feels harder to
breathe through the tube, he says restriction present, and otherwise
restriction absent. Whenever the experimenter says correct, no
changes in S2 occur, but when she says incorrect, the error e3 induces
a change in S2 equivalent to the subject saying to himself (for example)
Oh, now I see; restriction feels like _that_, not like I thought it did
before. In effect, the parameters of the neural network in S2 change
so that its output is the opposite of what it was the last time the
experimenter said incorrect.

Heres your original diagram for comparison:

Rick Marken (961003.0930)
                                                  r3 (zero error)
                                                  >
                                                  v
Subject r1 ("correct") p3-------->C3----------e3
                      > > >
                      v | |
               p1---->C1--------------e1 |
               > > >
               > r2 |
               > > >
               > v |
               > e2-------- C2<--------p2 |
               > > > >
               > > > >
               > > > >
               > v | |
______________|S1|_________|O|____________________|S2|<--------
Environment | "restriction present" "restricted"
               > "restriction absent" "unrestricted
               > >
               q v
            "correct" <-- |experimenter|
           "incorrect"

My problem is that I dont understand what C2 is doing, whether you call
r2 offset, criterion, or bias. What is it comparing? And how
does it affect the mapping between p2 and the output from O? Is it
inhibiting output from O while reorganization of S2 is going on? How
does the bias work?

Thanks for your help.

John

ยทยทยท

--
John E. Anderson
jander@unf.edu