[ACT-R-users] ACT-R Learning Mechanisms

[Dan Bothell]

--On Thursday, April 10, 2008 1:06 AM -0700 sima najafi <nj_sima at yahoo.com> 
wrote:

>
> Dear All,
> I have A question about ACT-R Learning Mechanisms.
> what i found till now has been the production compilation mechanism
> combining several existing rules into one. but i am looking for  a
> mechanism to learn rules based on facts in the declarative memory
> specially in the first phase where no rules exist.
> is there any mechanism?
> Thanks
>

There is no mechanism which can convert chunks into productions
directly.  However, the production compilation mechanism does
specialize the productions it creates based on chunks which
are retrieved.  When combining two productions, if there is a
retrieval request in the first production and that chunk is used
in the second production then the production created from those
two via production compilation incorporates the information from
that chunk and does not require the retrieval from declarative
memory.

If you have not worked through the ACT-R tutorial, I would suggest
looking at the demonstration model from unit 7.  The paired-learning
model in unit 7 shows how a very generic set of productions which are
capable of retrieving instructions from declarative memory can lead to
learning rules which are specific to a particular task.  That process
has been used to successfully model the learning of new tasks from
declarative information in a variety of different situations.

Dan