associative learning

[Christian Lebiere]

> My first problem:
> Because of continued creation of new chunks and the Prior Strength
> Equation S*ji = ln(m) - ln(n) states shown later are stronger associated
> with their components as compared to states shown earlier. That effect
> is due to the fact that initially there are only about 75 chunks in
> declarative memory and cannot be found in the data.
> - Is the assumption of the initial number of chunks <100 realistic?
>    (If it were 1000 chunks initially, the unwanted effect would almost
> disappear.)
> - Has anybody encountered a similar problem?
> - Are there solutions to circumvent the effect?

No.
Yes.
Yes.

There are probably many more than 100 chunks in your brain, so m should be
much larger than that to reflect the entire declarative memory (though it
is  usually unclear what the right value of n would be).  One fix that has
been used, especially in the list learning models, is to set the total
number of chunks (m) at some number.  This can be done with:

(setf *wme-number* 100)

However, ACT-R will keep adding or removing chunks, thus constantly
changing that number.  To prevent that from happening, you can define a
function that resets the number to the constant value and make that
function the value of one of the hook functions that is called at every
iteration, e.g. *cycle-hook-fn*.  This is the resulting code from the
Murdock model:

(defun reset-wme-number-murdock (instantiation)
  (declare (ignore instantiation))
  (setf *wme-number* 100))

(setf *cycle-hook-fn* #'reset-wme-number-murdock)
 
> - Can anybody explain me the sudden drop of IAs in the first cycle of
> the subgoal? (I cannot find the answer in the 1998 book.)

No idea.  I would have to try the actual model to be able to trace it.

> In the meantime I have found out something about activation sources:
> Typing "(activation-sources)" after firing a production that pops a
> subgoal returns the chunks of the last subgoal although the goal stack
> viewer tells me that the subgoal had already been popped. As far as I
> can see in the source-spread items of the declarative memory viewer this
> causes trouble in the spread of activation.
> - Is this a bug? (I use the ACT-R 4.0 Environment for Allegro CL 3)
> - If so, could it cause the sudden drop in the IAs?

It would be a bug (which might conceivably cause the IA problem), but I
cannot reproduce it using plain ACT-R 4.0.  I will take up the problem with
our PC-master.

Christian