sensitivity analysis, multilayer, feed-forward, back-propagation neural network using MATLAB.

How can I carry out a sensitivity analysis, that is, the effect of input parameters on the output of a multilayer, feed-forward, back-propagation neural network using MATLAB. What is the code for this, or is there any inbuilt function to carry out ?

Answer 1

[Greg Heath]

For a simple MIMO MLP, the I/O relationship is

 [ I N ] = size(x);
 [ O N ] = size(y);

 y = b2 + LW*tanh(b1+IW*x);

 Just take gradients with respect to whatever parameter component you are interested in.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Andualem alemu]

here I request you to give me a description on what the variable b2, LW, B1, and IW represent. this makes it clear. thank you

[Greg Heath]

Which has more hits when you search on

 b1 b2 IW LW

 The NEWSGROUP

 or 

 ANSWERS?

[Andualem alemu]

the answer you gave is the following for sensitivity analysis of MLP. From the formula you gave b1 b2 IW LW are not clear. What are they? [ I N ] = size(x); [ O N ] = size(y); y = b2 + LW*tanh(b1+IW*x); Just take gradients with respect to whatever parameter component you are interested in.

[Greg Heath]

I am confused. Yesterday you wrote

 " this makes it clear. thank you"

but today you wrote

 "...are not clear."

What has changed? More to the point: Did you learn anything at all from the formula? If so, what?

The following should help:

 h = tansig( b1 + IW*x ); % HIDDEN LAYER
 y = b2 + LW*h;           % OUTPUT LAYER

[Andualem alemu]

I am sorry that I am bothering you too much. But it is because I am absolute beginner and so curious to know, work on and understand MLP and ensemble MLP for time series forecasting!!

Dear sir my question is in the formula:

y = b2 + LW*tanh(b1+IW*x);

What is Y, b2, LW, b1, IW and x ?

 what are they representing and how can I find their value?