Scientific computing has come a long way in the last decade,
and the progress doesn't appear to be slowing down. As in
all walks of life these days, the computers used for
scientific simulations are concomitantly becoming more
powerful and less expensive. As a result, just about every
researcher, if he's so inclined, has the capability of
performing very high quality calculations in his field
of endeavor. Even with the wonderful breakthroughs of
recent years, however, there are still many types of problems
that just can't be satisfactorily handled through computer
simulation. In many of these cases, patience will continue
to be a virtue, as the inevitable growth in computing power
will enable researchers to study the problems they have
using existing methods. For other problems, however, it
may be necessary to shake things up a bit and move beyond the
confines of traditional computational methods. One promising
effort in this area is the growing focus on neural networks.
Before beginning to understand neural networks, it's probably
helpful to step back and look at some of the basics of
traditional computation. In general, computers are just
"dumb" boxes that do what humans tell them to do. Unlike most
people, computers are masters of following directions.
This is, of course, generally desirable, as the machine can
dedicate all of its resources to carrying out our instructions.
But such cold logic and strict adherence to the rules can often
leave a chasm between what the human intends and what the
machine actually performs. As well, such rigidity is frequently
the precise limitation on the ability of computational methods
to satisfactorily deal with a particular problem.
Indeed, systems like large-scale chemical processes and world economies, which are constantly evolving and don't necessarily have nice, neat conclusions, are often the subject of modeling efforts. In these cases, it would be of tremendous benefit to be able to employ a computer, or network of computers that could "learn" about the model and the forces that influence it as they evolve. This is the basic idea behind neural networks which are basically crude models of the human brain. Next time, we'll look at the basics of how such networks are constructed, and then we'll examine some of the potential applications of this exciting technology in the physical sciences.
Go To Page: 1
