Discrete Representations in Working Memory: Developmental,
Neuropsychological, and Computational Investigations
If you mentally rearrange your office furniture to
accommodate a bigger desk, you're using what's called working memory. This is
what the brain employs whenever a task requires multiple steps with
intermediate results that need to be kept in mind so that the task can be
completed. You rely on working memory, for example, when you calculate a tip in
a restaurant, or compare features of mid-size cars made by different companies,
or read a newspaper article. The project "Discrete Representations in Working
Memory: Developmental, Neuropsychological, and Computational Investigations"
undertook to explore how people learn to use working memory to assist in
The project, supported by the National Science Foundation
KDI initiative, was the collaborative effort of a team from the University of
Colorado at Boulder: Principal Investigator Akira Miyake, Professor of
Psychology; and co-PIs Yuko Munakata and Randall O'Reilly, both Associate
Professors of Psychology, and Michael Mozer, Professor of Computer Science.
In working memory, information is maintained in an active state
in order to direct what the brain is processing and can be updated when new
information comes in. Dr. Miyake and his team hypothesized that to hold
information in this accessible state, even in the face of delays and
interference, working memory representations should be discretemeaning a
finite amount of information should be represented in the brain. The team
welcomed the challenge of investigating this hypothesis. "We were all
interested in these general topics," said Dr. Munakata, "and we all were coming
at them from slightly different perspectives, so we were really excited about
working across our different perspectives."
They studied the ways in which the unique demands placed on
the working memory system shape how it learns and develops, and how this
affects the use of working memory as a whole. They did this in part with
behavioral studies on 3-year-old children, adults, and some adults with brain
damage. The team found that, for some experiments, they couldn't replicate the
results of previous studies that supported their hypothesis. This led them to
explore other waysboth experimental and using computational
modelsto test their theories.
The team developed a new computer modeling framework, one that
describes computation in the brain in terms of how information is transmitted
along processing pathways. According to Dr. O'Reilly, "What I did was develop
computational models that looked in more detail at the biological basis of how
these discrete representations might work in the cerebral cortex. Mike Mozer
was working on somewhat more abstract computational models that were in some
ways easier to understand but less closely tied to the biology, and then I
worked on models that were more closely tied to the biology and were more
complex. So it was really a nice combination of multiple levels of analysis."
The work of Drs. Miyake, Munakata, O'Reilly, and Mozer has
applications in a number of ways. First, the project contributes to the growing
body of literature that attempts to understand how different regions of the
brain are specialized for certain functions. "The hope is that people would
build on our work," said Dr. Munakata. "I know that that's already happening."
Their research is also useful for the many aspects of human activity that rely
on working memory, from military planning, to air traffic control, to economic
forecasting, and even to scientific research. It may also
contribute to knowledge of techniques for rehabilitating working memory that
has been impaired by brain injury and methods for assisting the development of
working memory in children. In addition, understanding the mechanism of human
learning and development can lead to an ability to design more intelligent and
adaptive computer systems.
Back to Top of Page