[From Chris Love (920909.15:30)]
[To: Anyone concerned with the control system imagination issue]
I've been *off the air* although I've been heavily involved in
the *Little baby's* development. I read Bill's post on imagination
[Bill Powers (920906)] and felt that something was either left un-
explained or missing. This dealt specifically with the objective(s)
of imagination. To me, an objective is to assist ECS' that are
having difficulty arranging their weights to percieve the desired
control variable (I think you guys call it CEV now). Now I like
Bill's proposal since it does away with having to *pair up* each
and every percept (input) with the corresponding *output* of an
ECS in the imagination loop. This is the way I originally
percieved this to occur. By Bill's proposal on imagination, I
mean that he uses a single signal, that signal being the processed
error signal of the ECS, which he feeds back into the comparator (is
that right Bill - this is how your diagram looked).
So what's my problem? The objective of imagination, in my view is
to better organize the control system. I mean if you didn't do
something to improve the operation of the system then why do it?
Would it be simply to know *if* the control system can control the
vble. or more specifically if an ECS (or group) can control their
So what's my proposal? Although I really like the idea of working
*only* with that 1-D error signal, it appears that imagination requires
the merging of each percept with the corresponding (matched) output.
The end result is the same; that result being that their is a closed
path through the ECS.
What's the difference (between Bill's proposal and this one)???
Well, first and most important, as I see things, is that an imrpovement
results from using imagination, as it does for us in most cases. For
instance, if I imagine (or in my own words ***picture***) how it would
be to throw a dart into the centre of a dart board, I do certain things
like feel the weight of the dart, use past experience about throwing,
test out the inertia of my arm, etc, and use all this information to
make the perfect toss. This preparation process is providing me with
references that I feel will bring about an improvement to my game.
In terms of the control system, these *improvements* occur when the
imagination loop is running and adjusting the weights (percept/output).
Now, what I'm not proposing yet, but will soon be investigating is
*HOW* this should be done. That's my next step. Sometimes, (this is
more a feeling) I see the effect of imagiantion as being similar (maybe
only in results) to the big discussion on random reorg.
This finishes my opinion on imagination.
Where am I in the Little baby? Well, I have just finished developing
the beautiful recurrsive lowpass filter (2nd order) that will be used
in the control system. Based on the Nyquist theorem, etc., each ECS
beginning at the lowest level, will have an ever decreasing bandwidth,
wich will always be half of its lower level neighbor. THis is also
nice from another perspective. This is to say that if you desire to
*skip* a layer, you will still be safe in terms of bandwdth/aliasing
since the highest layer into which the percept enters will always have
an adequate passband necessary for anti-aliasing AND their is a nice
*somewhat* linear phase delay in the filter too!!
So now the question arises as to how I will propose to test out these
theories? I now want to make a pseudo control system, where most
everything that can be preprocessed is, excluding a single ECS layer.
Here I can isolate the effects of the various learning/reorg. algorithms
that the group here at
DCIEM (Martin Taylor's group) will develop. I hope this will provide
a nice testbed for any proposals that come up on the net for percept
learning and/or output learning. I hope this to be able to provide
a fast and easy way of verifying their worth rather than disputing
them in theory. I think Bill will agree here as well as others.
AS they (who are they anyways???) say - "There's no substitute for the
REAL THING", huh!
Back to the books/Prograph (software).
Chris Love. (DCIEM)