ARTIFICIAL INTELLIGENCE
Computers On The Brain
Neural Networks
by Wolf von Rönik
Think neural networks are too difficult a concept to learn? They're
not. Take a look.
In a broad sense, neural nets are computer
models that attempt to mimic the "wetware" or actual physiological
structure and functioning of the brain (which is outlined in the sidebar,
"Neurons"). These models use the following analogies:
In the simple model displayed in Figure 1, I1, I2, and I3 represent
inputs. Moving down, each input reaches a weight (synapse) labeled W1,
W2, and W3 where input and corresponding weight are multiplied. This new
information is conducted (dendrite) as input to the node (soma). Inside
the node, the input-weight products are summed. This sum is then sent as
output along the single output (axon) arrow on the right to be compared
with a desired historical response. If the output and the historical response
agree (that is, a threshold is reached), you follow the lower arrow and
the model receives new input. If they don't agree, you return to the weights
via the top arrow, revalue the weights, and repeat the process until the
output and desired historical response agree.
This deceptively simple model allows a neural net to do something that
is impossible for an expert system: The net can "learn" to draw
conclusions from input of data rather than need to be told what conclusions
to draw in response to input. This amazing attribute is the result of the
net's ability to learn by altering its weights until it matches input with
desired historical response to a high degree of accuracy. In a neural net,
the weights are analogous to memory.
FIGURE 1: A SIMPLE NEURAL NET MODEL. The inputs are weighted
and multiplied. The input-weight products are summed and sent as output
(historical response). If the output and historical response agree, then
the model receives new output. If they don't agree, the weights are reevaluated
and the process is repeated until the output and desired historical response
agree.
...Continued in the April 2003 issue of Technical Analysis of STOCKS & COMMODITIES
Excerpted from an article originally published in the April 2003
issue of Technical Analysis of STOCKS & COMMODITIES magazine. All rights
reserved. © Copyright 2003, Technical Analysis, Inc.
Return to April 2003 Contents
