FW: imagination and memory

[from Jeff Vancouver 990726.1020 EST]

Attached: "budget4b.mdl" "Sample data.xls"

Marc, Bruce G., and whoever is interested:

I believe I have a protocol for testing memory models (e.g., alternative
modes). We can use the SimNurse research simulation to collect the real
person data to compare with a SD model. I am concerned that reviewers might
ask how the model (in this case PCT) accounts for how the memories got
there, but for the sake of putting reasonable boundaries on the project, I
think we might be able to make it an assumption. Few would argue that we
have memories. So this issue of learning (although it bothers me) can be
ignored for now.

My guess is that Dag's continuous model is more accurate then Bill's
switches model. We might be able to conduct sensitivity analysis on a key
parameter to support one over the other. But that is a detail we can worry
about later.

As far as the TEST, it has already been done for the budget goal. Of
course, it will need to be done for each participant, but the point is, we
can collect the data.

I have attached the SD model which matches data from several participants.
You can run the model and look at the graph. As Bill has noted, the SD model
has a serious flaw (in his opinion), which is that I am integrating 1 (in
the Trial level variable). The result is that I cannot properly model the
time lags (or change the lag variable). I figure that Marc can get passed
this problem (I do not consider it a serious problem because it relates to
my inability to reproduce the environment in the SD model, not my inability
to reproduce the system).

Anyway, to this model we need to add a layer regarding deleting shifts as a
lower-level unit (not hidden within the input function as it is now). I
also think we need to make the internal functions level variables.

As I mentioned, one can infer (based on memory), that deleting shifts
reduces the cost of the schedule without actually looking at the cost.

Any questions? (I am sure there will be some)

Jeff

budget4b.mdl (61 Bytes)

ยทยทยท

-----Original Message-----
From: Control Systems Group Network (CSGnet)
[mailto:CSGNET@POSTOFFICE.CSO.UIUC.EDU]On Behalf Of Marc Abrams
Sent: Friday, July 23, 1999 10:49 PM
To: CSGNET@POSTOFFICE.CSO.UIUC.EDU
Subject: Re: imagination and memory

from [ Marc Abrams (990722.2045) ]

[from Jeff Vancouver 990723.1000 EST]

I agree with 2-3 levels in model. I am less sure what you mean be
"assumptions," so it is hard to agree/disagree (former prefered).

The hierarchy is a proposed structure. When we talk about a specific Level
Relationships, for example ) we are assuming that a Level with the
characteristics described by Bill does in fact exist. When we talk about the
various modes Bill proposed in Chap 15 we need to make certain assumptions
about which modes are in effect at what levels with regard to a specific CV.

One assumption I would make is that all the levels modeled would be

"relatively

high" (e.g., program to category).

Yes. I initially envisioned 3 "general" levels. The first level does not
have a memory component ( as proposed by Bill ) The second level would
represent either singularly ( a specific Level ) or a combination of lower
Levels. With the third Level representing either singularly or a combination
of higher Levels. With refinement coming from the iterative modeling
process. Which would incluse our ability to aquire data and to validate the
model against actual observations.

What I am having trouble with conceptually is learning. Assuming a I am

correct that memory >stores a perception-perception link (where one is
higher than the other), how is that

link created. Can we model that? Or would that be an assumption (given a
link in memory, here is how imagination mode might work...)?

"Learning" is a _very_ braod concept and can mean a lot of things. For
instance, Are you talking about how new "Controlling Processes" are formed
reorganization ). Are you talking about how we utilize memory to set
reference levlels?, etc Can you elaborate on your ideas about your
perception-perception link and how it differs from Bill's proposed modes.
The "links" as proposed by Bill were by switches. Dag Forssell proposed a
continuous approach.

Yes, I believe we can model it. It's not only how "imagination mode" or
"automatic mode" "work" what is essential is understanding the effects these
modes have on our ability to control. By that I mean _what_ we control and
how we might maintain and change CV's.

_Everything_ we model initially will be _assumptions_. We are initially
assuming that a) a hierarchy does in fact exist or is plausible. b) the
modes Bill proposed are in fact plausible c) That we have the ability to
acquire the necessary data to validate what we are doing. This will _not_ be
easy. It is complex and we are working from a number of initial assumptions.
But I think the payoffs are huge _if_ we can help validate the existence of
a hierarchy, memory modes, and the interactions both might have, in our
ability to control.

Marc

[From Chris Cherpas (990727.0930 PT)]

Jeff Vancouver 990726.1020 EST--

My guess is that Dag's continuous model is more accurate then Bill's
switches model.

Does anyone know where I can view Dag's continuous model?

Best regards,
cc

from [ Marc Abrams (990727.1344) ]

[From Chris Cherpas (990727.0930 PT)]

Does anyone know where I can view Dag's continuous model?

Dag made a presentation at the 1994 CSG conference about it. It's on tape #1
available from Dag for a very nominal cost. Tom Bourbon, and Bill are also
on the tape. _Highly recommended_

There was also some discussion on the net prior to the conference about
memory. Anyone interested in those archived ( still not done :slight_smile: ) posts
should contact me and I would be happy to send them off to you.

The Control diagrams that Dag uses in his book on management "shows" the
memory connection but does not explain it.

Maybe Dag would do us the honor of coming on CSG and updating us on his
thoughts about memory in the control model. What do you say Dag?

Marc