Main Page Files Productions Problem Spaces Operators Goal Hierarchy Groups Prod. By Type All
Symbolic Concept Acquisition
Symbolic Concept Acqusition (SCA) is a model of concept learning
implemented in Soar. SCA was developed by Craig Miller. The
version of SCA documented here has been updated for Soar8 by Bob Wray. The minor modifications made to SCA for
compatibility with Soar8 are detailed in the documentation. This
documentation also includes a conceptual overview of SCA, as well
as specific documentation of the Soar production code.
SCA has also been extended
to include knowledge to map novel or unknown values of features to known values. Novel
features are often
introduced in transfer tasks in psychological experiments. The
solution proposed in this version of SCA is to allow novel
values to be mapped to known values. This mapping process competes
with abstraction. The result is
that unknown values may be ignored (via abstraction) or mapped to
a related value. (more information)
All the files needed for SCA are included in
sca.zip.
To provide a "working model",
some additional files are included
that can be used to run a simple model and trace the action of
SCA. These files are located in the directory test-harness/.
The test-harness productions set up a training instance for
prediction (no feedback), then learning (feedback provided), then
halt the model. You can step through the execution of this
example to see how the abstraction process works, and the
differences between prediction and training in the model.
Several logfiles are included (links below) that also
illustrate SCA with this test-harness.
Additional information about SCA:
- Miller, C. S. (1993). Modeling Concept Acquisition in the Context of a Unified Theory of Cognition. Ph.D. thesis, The University of Michigan. Also available as Technical Report CSE-TR-157-93.
- Miller, C. S. and Laird, J. E. (1996). Accounting for graded performance within a discrete search framework. Cognitive Science, 20, 499-537. (abstract)
- Wray, R. E., and Chong, R. S. (2003). Explorations of quantitative category learning with Symbolic Concept Acquisition. Presented at 5th International Conference on Cognitive Modeling (ICCM). Bamberg, Germany. April.
- A brief overview of SCA and its application to a transfer task
- sca.zip: zip file of SCA productions and this documentation
- An annotated trace of SCA on a single instance, showing prediction and training behavior, decisions, and production firings.
- Original watch level 5 trace showing all changes to WM during the run, along with decisions and production firings (not annotated).
Documented at line
1
of file
SCA_source.soar
Generated with SoarDoc on Fri Apr 25 18:21:37 2003 by wray