# How to implement neural network?

Status
Not open for further replies.

#### SHARMILAK27

##### Newbie level 4
ppppplssss tell me about neural network for number recognition ?
feedforward backpropagation neural network
how to apply input ,my digit image is of 11*11 size ,
then how much neurons needed in input layer,(1 pixel/neuron or 11 i.e column/neuron)
pls help

#### andre_luis

##### Super Moderator
Staff member
11x11 inputs at each neuron.
Number of neurons, defined by number of characters to be recognized.
The choice of pattern is defined by output that reachs maximum value.

MathWorks have a good tutorial. See also :
Image Recognition with Neural Networks HowTo

+++

Last edited:

#### SHARMILAK27

##### Newbie level 4
11x11 inputs at each neuron.
Number of neurons, defined by number of characters to be recognized.
The choice of pattern is defined by output that reachs maximum value.

MathWorks have a good tutorial. See also :
Image Recognition with Neural Networks HowTo

+++

many thanks for reply
I want to recognize 6 digit at a time bz project is printed pincode recognition, so according to ur reply I need 6 image of 11*11 size to each input neuron and 10 neuron for hidden and 1o for output layer,
m I right?
pls reply with much clear n extra info which will needed to my project,
thank u

#### andre_luis

##### Super Moderator
Staff member
You need 1 output for each digit.
The choice of amount internal layers don´t follows a precise rule, but have to try empirically.
In practice, is noticed that number of hidden layers must be around : 2/3 Input layer size + output layer size.
It menas to above case, you need :
INPUT LAYERS : 11x11 = 121
HIDDEN LAYERS : 121*(2/3) + 6 = 80 + 6 = 86
OUTPUT LAYERS : 6

+++

Last edited:

#### SHARMILAK27

##### Newbie level 4
You need 1 output for each digit.
The choice of amount internal layers don´t follows a precise rule, but have to try empirically.
In practice, is noticed that number of hidden layers must be around : 2/3 Input layer size + output layer size.
It menas to above case, you need :

+++

thank u very much
this is too useful answer
I hv read details of neural network n i hv tried to xor problem but i observed its output gets change at each time of execution.
why is this?
I mean, without any change in code sometime goal met,sometime not.
I want to know how to make stable it i.e output.
pls reply thank u:?::?::?:

#### andre_luis

##### Super Moderator
Staff member
Is needed more details to perform better analysis.
I´m not sure if I understood the problem.

--> What tool are you using to test ?
--> Topology circuit ?
--> Data entry ?
--> etc...

+++

Status
Not open for further replies.