ACT-R Parallelism

[Troy Kelley]

Mike Byrne <byrne at rice.edu> on 04/09/2001 12:25:23 AM





To:   Troy Kelley <tkelley at ARL.ARMY.MIL>
cc:   act-r-users at andrew.cmu.edu
Subject:  Re: ACT-R Parallelism


On 01.04.06, Troy Kelley <tkelley at arl.army.mil> wrote:

> I seem to be a little confused about the parallel aspect of ACT-R
> P/M.  I am working with visual objects and the documentation for
> ACT-R P/M says that, "the vison module checks the icon to see if
> there is a feature in the icon that matches the chunk requested
> by ACT-R.  If one existes, a chunk representing that visual
> location is created and returned to ACT-R in zero time."  I guess
> I also assumed that all chunks returned by ACT-R P/M in zero time
> would have the same activation levels, but as far as I can tell,
> they don't.

-It is correct that not all chunks of type VISUAL-LOCATION will have
-the same activation.  This will be a function of what the model is
-currently attending to, because the slots of the currently-attended
-visual object are activation sources.  Thus the VISUAL-LOCATION
-chunk associated with that object will have a higher activation
-than other VISUAL-LOCATION chunks, unless they happen to be stuffed
-into the goal.

> So... if they don't have the same activation levels, then they
> really are not returned in parallel, because some are more active
> than others.

-I'm sorry, but I don't understand your logic here.  Differential
-activation of VISUAL-LOCATION chunks doesn't affect the behavior of
-the Vision Module one iota.  That is, the activation level of a
-VISUAL-LOCATION doesn't affect how rapidly it will be returned with
-a TIME NOW test.  That time is always zero, regardless of the
-chunk's activation.

I understand this.. I guess I was using activation as a measure of
parallelism and not just time.


-For example, consider the following production LHS, and assume
-there are two objects on the screen, one with an x-coordinate of
-500 and one with an x-coordinate of 150:

-=goal>
-   isa          example
-=loc1>
-   isa          visual-location
-   time         now
-   screen-x     150
-=loc2>
-   isa          visual-location
-   time         now
-   screen-x     500
-
-This production would find (or create) two VISUAL-LOCATION chunks
-and the total matching time for this production would be zero.
-This qualifies as parallel in my book, and will happen irrespective
-of the activations of those chunks, which could be different
-depending on what's in the goal and what VISUAL-OBJECT is currently
-attended.
-
-Now, if you are retrieving a VISUAL-LOCATION chunk from memory
-(i.e. doing it *without* a TIME NOW specifier), then you are
-correct, they will be retrieved serially and will be affected by
-having differential activation levels.  If you omit the TIME NOW
-specifier, then RPM never gets called; you are simply requesting a
-pair of retrievals from declarative memory, which works the way it
-always has.
-
> If they were returned in parallel, they all should have the same
> activations.

-This is simply not true; see my explanation above.

I understand your explanation, it just seems to me that if ACT-R P/M
perceives multiple items in parallel, the subsequent chunks created by that
perception should then have the same time AND activation levels associated
with them.


> So.. do I have to do a match on all the portions of the screen
> that I want returned as chunks with the same activation?
> Couldn't I do that without ACT-R P/M?

-Again, I apologize, but I don't understand what you want here.
-Could you clarify or re-ask this question?  I suspect some code
-would *really* help here.

Actually, the code you present above is exactly what I was talking about
here.  But again, as you point out above, developing the code in this way
will not return chunks with the same activations for the reasons you note
above.

I got it.. thanks.  I guess I need to re-think how I want to construct my
model.

Troy