Memory - Computational Explorations in Cognitive Neuroscience - page 315

tion (e.g., the AB-AC list learning task) causes prob-
lems for our generic posterior-cortical model because
of its overlapping, distributed representations .The
hippocampus can be seen as a complementary memory
system that uses sparse, pattern separated represen-
tations to avoid this interference. By using two special-
ized systems, the benefits of each are still available —
overlapping distributed representations in the cortex can
be used for inferential processing of semantic infor-
mation, while the sparse pattern separated representa-
tions of the hippocampus are used for rapidly encoding
arbitrary information.
In the domain of activation-based memory, the
unique demands of rapid updating and robust main-
tenance imposed by many working memory tasks re-
quires specialized neural mechanisms that appear to
be characteristic of the prefrontal cortex, and are also
in conflict with overlapping distributed representations
characteristic of the posterior cortex, which tend to
cause excessive spreading activation .
We saw how each of the memory components con-
tributes to specific memory phenomena. The poste-
rior cortex supports a relatively simple form of memory
called priming , where prior processing results in the
facilitation of subsequent processing. Priming has both
weight-based ( long-term ) and activation based ( short-
term ) aspects. The hippocampus supports relatively
interference-free rapid arbitrary learning, as in the AB-
AC list learning task. The prefrontal cortex can learn to
dynamically control its robust, rapidly updatable active
maintenance in a working memory task. The memory
components can also interact in a wide range of other
memory phenomena.
We find it notable that the cognitive architecture
can be differentiated according to memory — because
memory is so tightly integrated with processing in a
neural network, it makes sense that qualitative distinc-
tions in memory representations underlie the distinc-
tions between different types of processing in the cog-
nitive architecture.

Squire (1992) and Cohen and Eichenbaum (1993)
have been influential and cover a lot of data on hip-
pocampal function.
McClelland and Chappell (1998) presents a unified
neural-network based treatment of a number of human
memory phenomena.
O'Reilly and Rudy (in press) presents a detailed treat-
ment and models of the role of the hippocampus and
cortex in animal learning phenomena.
Miller et al. (1996) gives a very detailed treatment of
influential studies on the neurophysiology of activation-
based memory in the frontal cortex.
Miyake and Shah (1999) provide a very recent col-
lection of articles on working memory, mostly from the
human behavioral perspective.

9.9 Further Reading

Schacter (1996) provides a very readable introduction
to the cognitive neuroscience of memory.