Re: FIGHT! FIGHT! FIGHT!

Roy Lewallen wrote:
I'm sure that somewhere in one of your texts you can find the definition of linear as applied to networks. Once you do, though, a little thought is required to discover that y = mx + b doesn't satisfy the criteria for network linearity.

To be linear, a network has to satisfy superposition. This means that:

If y1 is the response to excitation x1 and y2 is the response to
excitation x2, then the response to x1 + x2 must be y1 + y2.

Let's try that with your function.

The response to x1 is:

y(x1) = mx1 + b

The response to x2 is:

y(x2) = mx2 + b

The sum of y(x1) and y(x2) is:

y(x1) + y(x2) = m(x1 + x2) + 2b

But response to x1 + x2 is:

y(x1 + x2) = m(x1 + x2) + b

These are not equal as they must be to satisfy superposition and therefore the requirements for linearity.

Roy Lewallen, W7EL

Richard Harrison wrote:
Roy Lewallen, W7EL wrote:
"But of course you realize that the function y = mx + b doesn`t meet the
requirements of a linear function when applied to network theory."

Works for me.

Linear means the graph of the function is a straight line.

f(x) = y = mx + b is called linear because its graph is a straight line.

A straight line is the shortest distance between two points.

In y = mx + b, m is a constant determining the slope of the line. x is
is the independent variable. b is the offset or point along the x-axis
where the line crosses.

y then is a linear function of x because its slope is always mx, but
displaced in the x-direction by a constant value, namely b.

y is linear the same as IR is linear, or by substitution, E is linear in
Ohm`s law where E=IR. For any value of I, voltage = IR and the graph of
I versus E is a straight line with a slope equal to R.

Resistance is a common factor in network theory.

Best regards, Richard Harrison, KB5WZI

Not that it means anything, but the linearity requirement is met when
b = 0, which, of course, is a subset of the family of equations of the form y = mx + b.

73,

Chuck
NT3G
.

Relevant Pages

  • Re: MIMO system in NN Toolbox
    ... But since the network is trained, ... ALWAYS try the constant and linear models first ... DIVIDEVEC command below) is so much larger, ... nontraining validation set. ...
    (comp.soft-sys.matlab)
  • Re: FIGHT! FIGHT! FIGHT!
    ... I'm sure that somewhere in one of your texts you can find the definition of linear as applied to networks. ... a network has to satisfy superposition. ... If y1 is the response to excitation x1 and y2 is the response to ... Linear means the graph of the function is a straight line. ...
    (rec.radio.amateur.antenna)
  • Re: Neural Network
    ... the neural network can not find good fitting for these data. ... At this stage I would feel confident removing input variables, ... Before designing a 12-H-31 MLP, take a look at a linear ... insuring weight estimate accuracy but is independent ...
    (comp.soft-sys.matlab)
  • Re: Neural Network
    ... the neural network can not find good fitting for these data. ... At this stage I would feel confident removing input variables, ... Before designing a 12-H-31 MLP, take a look at a linear ... insuring weight estimate accuracy but is independent ...
    (comp.soft-sys.matlab)
  • Re: Neural Network
    ... the neural network can not find good fitting for these data. ... coefficients among input variables 3 to 7 exceed 0.98. ... In fact when predictors are linearly dependent (or nearly ... it is usually wise to remove near linear ...
    (comp.soft-sys.matlab)