|
This first chapter works as a general introduction to Ballard's book, presenting
the principal motivation and some basic concepts. It starts by briefly reviewing the
mammals brain in
a hierarchical fashion (from subsystems to neurons, passing through maps). Next, basic principles in
computational theory,
including Turing machines and
complexity, are outlined and exemplified. The next
section motivates and introduces some of the crucial concepts in natural computation, including minimum
description length, learning, and architectures. The second example illustrating the former of these concepts
regards the receptive field of a neuron,
which is defined as the set of respective synaptic weights.
It should be observed that, in neuroscience, receptive fields are typically defined as the region of the
input space which can affect the activity of the neuron
[Goldstein
]. As a matter of fact, neither
of these definitions is quite accurate. While the more traditional definition, which is more geometrical,
fails to specify the intensity of the modulatory input influences of the input over the neural activity,
Ballard's alternative definition does not specify the spatial positioning and arrangement of the input
and synapses.
|
|
The penultimate section in this chapter presents an overview of the main objectives
and general organization of the book. Though Ballard observes that the "overall perspective of this book
is that learning algorithms develop behavioral programs", it is not immediately clear what is meant by
"behavioral", "algorithms", and "programs", but this becomes clearer as the reader proceeds further into
the book and re-read some sections. Such a feature is characteristic of many passages in this book.
The last section in this chapter comments on how the great challenge of using computers as a means of
better understanding the brain have met very limited success thus far. Ballards conclude this section
and chapter by observing that the goal of his book is to further this process.
|
|
|