Page 94 - Anatomy of a Robot

03_200256_CH03/Bergren 4/17/03 12:27 PM Page 79
COMPUTER HARDWARE 79
More information about analog computers can be found at www.science.uva.nl/fac-
ulteit/museum/AnalogComputers.html and at www.play-hookey.com/analog. The
Analog Computer Museum, dedicated to the history of analog computers, is at
http://dcoward.best.vwh.net/analog/.
Neural Networks
One of the finest computational engines known to exist is the human brain. It can solve
most complex, real-world problems much faster than a general-purpose computer,
albeit with less precision. Electronic computers are best suited to problems requiring
arithmetic capability and blinding execution speed, such as forecasting the weather. But
they are not good at solving problems requiring judgment or experience. The human
brain has the experience and “wiring” to take on problems that it has never seen before
and to solve them with speed and reliability. The parents of teenagers might argue with
this last statement, but they have never tried to live with a teenage robot struggling with
its computer’s programming so it can survive puppy love. Be assured, parents would
rather deal with a human teenager who, believe it or not, has amazing abilities com-
pared to a computerized robot.
So what is a neural network? Ever since humans first grasped the structure and pur-
pose of the human brain, they have dreamed of building an artificial brain. Many
designs for such a brain have been put forth, including neural networks. First, let’s look
at the human brain.
Brain cells, called neurons, are connected together in a vast array of tissue within the
brain. They communicate electronically with one another over neural connections called
synapses. This allows neurons to exchange information with nearby neighbors. Neurons
retain information (dubbed memory) chemically and electrically within the cell body
(see Figure 3-3).
The memory of a specific spring day, for example, might be spread out over a vast
array of neurons, which govern smell, sight, hearing, motion, and so on. The memory
of the spring day is distributed throughout the brain. Memories can be imperfect and
they can fade as individual neurons begin to lose their individual memory of the day.
Memories are stored almost like a photo spread out over the fabric of the brain. Neurons
might store more than one memory at the same time. This is why the remembrance of
one thing, like a spring day, might evoke the memory of another experience, like the
ice-cold water of a stream. A human, prodded to remember the spring day with the noise
of a brook, would likely dredge up the memory of stepping into a noisy, icy brook. The
fact that noise was in both memories ties the memories together. The human has learned
to be suspicious of brooks on spring days; they might be icy.
Learning is something general-purpose computers are not good at. Some neural-
network computers are designed to mimic the learning ability of the human brain. They