[From Bill Powers (931106.2020 MST)]
Martin Taylor (931106.2045) --
The Lang-Ham system behavior which you plotted is the behavior
without disturbances; with disturbances, the feedforward will be
incorrect and the negative feedback part will have to make a
correction.
The 1955 reprint, which I dug out of a pile, shows a system much
like what I proposed (I probably absorbed it without realizing).
However, they insert a function between the reference signal and
the comparator (which is probably why I discarded it), so the
_effective_ reference signal is no longer the direct standard
against which the perceptual signal is compared. The engineer,
standing off to one side, has to adjust this function and the
others in the circuit to achieve the desired relationship between
the reference signal and the OUTPUT. As usual, the system is
thought of as controlling output. If you make the B function a
straight-through connection you have my proposed hybrid system,
and the system will control its perceptual signal, making it
match the reference signal.
(P.S. In your recollected labels, A and B are interchanged. B is
the function between the reference signal and the comparator).
Tom's point, I think, is that for the kinds of behavior we have
modeled so far, adding this complication would gain us very
little by way of predictability. However, when we get around to
testing control when the feedback information is periodically
blanked out, my new hybrid scheme, similar to the Lang-Ham
scheme, will be required. The Lang-Ham scheme will not work when
the feedback signal is cut off, however, because it doesn't use
the one-way properties of neural systems to make the error go to
zero when the feedback signal (both ways) is lost.
ยทยทยท
----------------------------------------------------------------
The copyright notice is for the source code only. The executable
code can be used freely for the specific purposes of the sleep
experiment. No big deal in this particular instance. Consider
yourself a beta-test site, doing us a favor.
-------------------------------------------------------------
Norm Holland (931106) --
Surely a big part of your identity theme, Bill, is "control."
No doubt. What's the alternative? To have a reference-identity
and not act to maintain your perceived identify close to it?
Incidentally, is "HPCT" Hierarchical Perceptual Control Theory?
Yep.
With respect to the feedforward and feedback discussion, is it
possible to reformulate Hans' "ballistic" and "feedforward" as
feedbacks in which the reference signal is a prediction?
How do you mean, it "is" a prediction? A reference signal is a
specification against which the current perception is compared,
with the resulting difference driving behavior. Do you mean that
the reference signal is set as the result of some other system's
making a prediction? That really isn't necessary, because if the
control systems are working properly, the resulting perceptual
signal _WILL_ match the reference signal: no prediction about it.
In the open-loop case, the term "reference" is inappropriate,
because the outcome is not being adjusted on the basis of a
comparison of the outcome with any reference condition. The
output action is simply varied blindly as a function of an input
signal, with the eventual outcome being whatever it is.
Disturbances or changes in environmental properties do not result
in changes of output that keep the outcome the same.
Metaphorically, you could say that the ballistic system acts as
if it is predicting that no disturbances will occur, but of
course it's not necessary for any such prediction actually to be
made in order for the ballistic system to act as it does. It just
acts the way it's hooked up to act.
--------------------------------------------------------------
Best to all,
Bill P.