Introduction and Overview - Computational Explorations in Cognitive Neuroscience - page 19

to update the hypothesis next time around. Although
this has a more discrete flavor, we find that it can best
be implemented using the same kinds of graded neu-
ral mechanisms as the other kinds of learning (more on
this in chapter 11). Another more discrete kind of learn-
ing is associated with the “memorization” of particular
discrete facts or events. It appears that the brain has a
specialized area that is particularly good at this kind of
learning (called the hippocampus ), which has properties
that give its learning a more discrete character. We will
discuss this type of learning further in chapter 9.

in which we are forced to make a choice are presented
as such, and relevant arguments and data are presented.
We strive above all to paint a coherent and clear pic-
ture at a pace that moves along rapidly enough to main-
tain the interest (and fit within the working memory
span) of the reader. As the frames of a movie must fol-
low in rapid enough succession to enable the viewer to
perceive motion, the ideas in this topic must proceed
cleanly and rapidly from neurobiology to cognition for
the coherence of the overall picture to emerge, instead
of leaving the reader swimming in a sea of unrelated
facts.
Among the many tradeoffs we must make in accom-
plishing our goals, one is that we cannot cover much of
the large space of existing neural network algorithms.
Fortunately, numerous other texts cover a range of com-
putational algorithms, and we provide references for the
interested reader to pursue. Many such algorithms are
variants on ideas covered here, but others represent dis-
tinct frameworks that may potentially provide impor-
tant principles for cognition and/or neurobiology. As
we said before, it would be a mistake to conclude that
the principles we focus on are in any way considered
final and immutable — they are inevitably just a rough
draft that covers the domain to some level of satisfaction
at the present time.
As the historical context (section 1.3) and overview
of our approach (section 1.4) sections made clear, the
Leabra algorithm used in this topic incorporates many
of the important ideas that have shaped the history of
neural network algorithm development. Throughout the
topic, these principles are introduced in as simple and
clear a manner as possible, making explicit the histor-
ical development of the ideas. When we implement
and explore these ideas through simulations, the Leabra
implementation is used for coherence and consistency.
Thus, readers acquire a knowledge of many of the stan-
dard algorithms from a unified and integrated perspec-
tive, which helps to understand their relationship to one
another. Meanwhile, readers avoid the difficulties of
learning to work with the various implementations of
all these different algorithms, in favor of investing ef-
fort into fully understanding one integrated algorithm
at a practical hands-on level. Only algebra and simple
calculus concepts, which are reviewed where necessary,

1.7 Organization of the Topic

This topic is based on a relatively small and coherent set
of mechanistic principles, which are introduced in part I
of the text, and then applied in part II to understanding
a range of different cognitive phenomena. These prin-
ciples are implemented in the Leabra algorithm for the
exploration simulations. These explorations are woven
throughout the chapters where the issues they address
are discussed, and form an integral part of the text. To
allow readers to get as much as possible out of the topic
without doing the simulations, we have included many
figures and have carefully separated the procedural as-
pects from the content using special typesetting.
Because this topic emphasizes the linkages and in-
teractions between biology, computational principles,
and a wide variety of human cognitive phenomena, we
cannot provide exhaustive detail on all potentially rel-
evant aspects of neuroscience, computation, or cogni-
tion. We do attempt to provide references for deeper
exploration, however. Relatedly, all of the existing sup-
porting arguments and details are not presented for each
idea in this topic, because in many cases the student
would likely find this tedious and relatively uninfor-
mative. Thus, we expect that expert neuroscientists,
computational/mathematical researchers, and cognitive
psychologists may find this topic insufficiently detailed
in their area of expertise. Nevertheless, we provide a
framework that spans these areas and is consistent with
well-established facts in each domain.
Thus, the topic should provide a useful means for ex-
perts in these various domains to bridge their knowl-
edge into the other domains. Areas of current debate