developerWorks: Weave a Neural Net with Python
Jun 21, 2004, 07:00 (0 Talkback[s])
(Other stories by Andrew L. Blais)
"Hot things cool, obviously. The house gets messy,
frustratingly. In much the same way, messages are distorted.
Short-term strategies for reversing these things include,
respectively, reheating, cleaning, and the Hopfield net. This
article introduces you to the last of these three, an algorithm
that can, within certain parameters, undo noise. A very simple
Python implementation, net.py, will show you how its basic parts
fit together, and why a Hopfield net can sometimes retrieve a
pattern from its distortion. This implementation, while limited,
will still enable you to perform a number of instructive and
revealing experiments on the Hopfield net.
"I assume that you are reading this because you have some
computational problem. You have been advised that some neural net
algorithm might provide a solution. Specifically, the suggestion is
that you might employ a Hopfield net. I suppose further that you
need a precis of the idea so you can decide whether the suggestion
is feasible and warrants more exploration. The following broadly
sketched application of the Hopfield net may get you started
solving your problem..."