Entry

Computer:Artificial intelligence: Neural network: Simple: How create simple neural network? [Delphi]

Mar 13th, 2005 02:05
Knud van Eeden,

----------------------------------------------------------------------
--- Knud van Eeden --- 17 November 2003 - 00:28 am -------------------

Computer:Artificial intelligence: Neural network: Simple: How create 
simple neural network? [Delphi]

---

The purpose is here to get the basic idea, so that you can use
it to connect two or more of this basic units together.

---

The most basic unit of a neural network could be the following:

1. 1 input

2. 1 weight

3. 1 treshold

4. 1 activation function

---

This might be represented like the following:


        weight1 +---------------+     +--------------------+
 input1---->----|linear combiner|-->--|activation function1|->-output1
                +---------------+     +--------------------+
                                       ^
                                       |
                                       treshold

---

To create a very simple computer program calculating this,
you have basically 3 steps:

---

Steps: Overview:

Given

   1. activationfunction1

   2. input1

   3. weight1

   4. treshold1

1. Calculate what comes out of the linear combiner

   newsum1 = oldsum1 + input1 times weigth1

2. Calculate the activation function

   x1 = newsum1

   y1 = activationfunction1 as a function of this x1

3. Calculate the output

   if y1 is greater than given treshold
   then fire else do nothing

---

Steps: Worked out:

Given:

    activationfunction1 = 1 / ( 1 + exp( - x1 ) )

    input1 = 0.7

    weight1 = 0.5

    treshold1 = 0.2

sum1 = 0

1. Calculate what comes out of the linear combiner

    sum1 = sum1 + input1 * weight1

2. Calculate the activation function

    x1 = sum1

    y1 = 1 / ( 1 + exp( - x1 ) )

3. Calculate the output

    IF ( y1 > treshold1 ) BEGIN
     writeline( 'fired' );
    END
    ELSE
     writeline( 'do nothing' );
    END;

---

So working out this numeric example:

---

input1 := 0.7;

weight1 := 0.5;

treshold1 := 0.2;

sum1 := 0;

// 1. Calculate what comes out of the linear combiner

sum1 := sum1 + input1 * weight1;

// sum1 = 0 + 0.7 * 0.5
// or thus
// sum1 = 0.35

// 2. Calculate the activation function

x1 := sum1;

// or thus
// x1 = 0.35

y1 := 1 + ( 1 + exp( - x1 ) );

// or thus
// y1 = 1 + ( 1 + exp( - 0.35 ) )
// or thus
// y1 = 0.58

// 3. Calculate the output

if ( y1 > treshold1 ) then begin
 writeline( 'fired' );
end
else
 writeline( 'do nothing' );
end;

// now 0.58 is greater than 0.20 thus it will fire

---

Writing a small Delphi program to test this idea:

Steps: Overview:

 1. -Open Delphi

 2. -Create a new application

 3. -Put a button on the form

 4. -Double click on the button
     and fill in the following source code

--- cut here ---------------------------------------------------------

procedure TForm1.Button1Click(Sender: TObject);

var input1 : double;

var weight1 : double;

var treshold1 : double;

var sum1 : double;

var x1 : double;

var y1 : double;

begin

input1 := 0.7;

weight1 := 0.5;

treshold1 := 0.2;

sum1 := 0;

// 1. Calculate what comes out of the linear combiner

sum1 := sum1 + input1 * weight1;

// sum1 = 0 + 0.7 * 0.5

// or thus

// sum1 = 0.35

// 2. Calculate the activation function

x1 := sum1;

// or thus

// x1 = 0.35

y1 := 1 + ( 1 + exp( - x1 ) );

// or thus

// y1 = 1 + ( 1 + exp( - 0.35 ) )

// or thus

// y1 = 0.58

// 3. Calculate the output

if ( y1 > treshold1 ) then begin

 ShowMessage( 'fired' );

end

else begin

 ShowMessage( 'do nothing' );

end;

// now 0.58 is greater than 0.20 thus it will fire

end;

--- cut here ---------------------------------------------------------

 5. -If you run this program, it will show a message box
     with 'fired'

---
---

Internet: see also:

---

Neural network: Link: Can you give an overview of links?
http://www.faqts.com/knowledge_base/view.phtml/aid/34415/fid/1760

----------------------------------------------------------------------