[ACT-R-users] Has 'Conceptual Knowledge' been modeled?

db30 at andrew.cmu.edu db30 at andrew.cmu.edu
Wed Sep 3 10:04:41 EDT 2014



--On Tuesday, September 02, 2014 3:29 PM -0400 "Sandra L. Vaughan" 
<sandralvaughan at gmail.com> wrote:

>
>
> ACT-R Group,
>  
> I need to either confirm that there are no existsing models of "Conceptual
> Knowledge" (see definition below) in ACT-R (or any other Cognitive Modeling
> Architecture), or find them if they exist.
>
> I have accomplished a rather exhaustive search of avaiable literature,
> including the ACT-R acrhives,and have not found anything.  So I thought I
> would send a request out to the group.
>  
> Thank you in advance for your reply.
>  
> Definition - conceptual knowledge
>
> \When we store experience in memory, we do not record every detail, as a
> physical recording would. We keep some of the information and drop other
> [perceived as unimportant] details. We can abstract from specific experiences
> to general categories of the properties of that class of experiences. This
> sort of abstraction creates conceptual knowledge involving categories: for
> example, chairs and dogs (p.154) [2]"
>
> [2] Anderson, J. R., Cognitive psychology and its implications, Macmillan,
> sixth ed.,
>
> 2005.
>


I think that many of the models which use an "instance based learning"
approach have that character to them, but probably don't refer to it
as "conceptual knowledge".  While they are usually more focused on a
specific task than learning something general like dog, they often
require only recording a subset of information from an experience
for later use and/or abstracting detailed past experiences into more
general "categories".  The assignment task in unit 5 of the ACT-R
tutorial requires explicitly paring down the available information from
a game in the model so that it can learn to play better by retrieving
from among those simplified past experiences.  Whereas instance based
learning models which use the blending mechanism can dynamically
generalize over detailed past experiences using features determined
during the task to create such knowledge.  I don't know of specific
references to models of that nature which you could get, but others
here can probably provide some.

Hope that helps,
Dan




More information about the ACT-R-users mailing list