Professor
Metcalfe's presentation was concerned with the concept
of binding in memory. Binding is a term used by cognitive
psychologists to describe the process of organizing
lower order elements into higher order representations.
Evidence from other studies has suggested that binding
occurs in one type of memory but not in another. These
two systems of the memory are referred to as explicit
memory and implicit memory. Explicit memory is thought
to involve conscious, controlled recollective processes,
whereas implicit memory is thought to involve unconscious,
automatic processes.
Professor
Metcalfe presented research on face recognition in normal
subjects. She compared her empirical results with computer
simulations of human memory. Earlier research suggested
that human subjects have trouble with faces which are
mixtures of previously studied individual features such
as eyes, nose, hair, and mouth. Subjects tend to recognize
these types of faces, referred to as conjoined faces,
as ones they had studied before when in fact they had
not seen them at all. Apparently, changing the individual
features of faces causes subjects
to confuse these faces with previously studied faces
which have not had their individual features changed.
One reason may be that they invoke the higher order
representations of faces which were studied previously.
What would
happen if a face mimicked the higher order representation
of a face even better than a conjoined face? This effect
was produced by superimposing any two faces that subjects
had studied previously to produce what Professor Metcalfe
referred to as a superimposed composite face. Professor
Metcalfe's research indicated that these faces, like
conjoined faces, were easily mistaken for previously
studies faces.
What happens
when the two types of face-mixing , i.e. conjoined and
composite faces, are compared? Which type of face invokes
the higher order representation of a stored face in
memory better? It would make the most sense for the
composite faces to be most easily mistaken for unaltered
faces studied before because, as discussed, they mimicked
the higher order representation of a face the best.
Professor Metcalfe's research indicated that this was
in fact true: subjects confuse the superimposed faces
more frequently with previously
studied faces than they do with conjoined faces and
previously studied faces.
What processes
underlie this process of binding the eyes, nose, mouth,
and hair into higher order representations? Computational
models of memory provide very specific equations for
how memory works and therefore can possibly reveal some
of the processes that underlie this ability. These models
have proved successful in the past in accounting for
a large variety of data on human memory and learning.
After inputting the faces into the four different computational
models, it was found that the models that best fit the
data were models that had the ability to convolve, or
autoassociate, features into a larger whole. The models
that did not have this additional feature were unable
to provide the results that were obtained with human
subjects reported earlier here.
Another feature
which proved essential in producing results similar
to the results obtained with the human subjects was
the capacity to store individual features all together
in a single storage box, which computational memory
designers refer to as a memory vector . Those
models that lacked this ability and designated that
items be stored individualy could not provide the data
Professor Metcalfe obtained. The models that fit the
data obtained from research on human subjects best had
the dual capacity to convolve, or autoassociate, individual
features from objects together and also had the ability
to store these combined features in a single storage
box.
Earlier findings
have suggested that binding is a process that occurs
in explicit memory. This is the area from which Professor
Metcalfe's research took its lead. By studying the effect
of randomly mixing features of a face such as the nose,
eyes, mouth and hair and comparing this with subject's
recognition of faces that were superimposed on each
other, it was found that binding is a process that involves
the formation of higher order representations of faces
such that the closer a face is to this higher order
representation, the more easily it would be mistaken
for this representation. Professor Metcalfe's research
also revealed, through the use of computational models
of memory, that binding is a process that involves a
combination of these features in a concrete manner and
in a single storage box or memory vector.
