Scribd
Upload
111 views

MLP

The document describes using MATLAB to simulate a multi-layer perceptron (MLP) neural network to model a nonlinear function. It includes: 1) defining training and test input/output patterns, 2) creating a 3-layer MLP architecture with different transfer functions, and 3) using MATLAB functions like newff, train, and sim to train the network on 100 epochs and test it on new input patterns. Figures show the training data, training progress, and test output matching the desired function.

Uploaded by

Parmalik Kumar
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
111 views

MLP

The document describes using MATLAB to simulate a multi-layer perceptron (MLP) neural network to model a nonlinear function. It includes: 1) defining training and test input/output patterns, 2) creating a 3-layer MLP architecture with different transfer functions, and 3) using MATLAB functions like newff, train, and sim to train the network on 100 epochs and test it on new input patterns. Figures show the training data, training progress, and test output matching the desired function.

Uploaded by

Parmalik Kumar
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

ME 599-5 ARTIFICIAL NEURAL NETWORK

: THEORY AND APPLICATIONS


Professor : Hyungsuck Cho

MLP Neural Network simulation using Matlab


The process to train and test a designed MLP neural network :
1) We make training patterns and test patterns.
2) A network architecture should be defined by newff MATLAB function with the
number of layers, neurons and transfer functions.
3) The defined neural network architecture is trained by train MATLAB function with
input patterns and training parameters.
4) We can easily check the result by using a sim MATLAB function.

Example :
1. Training Patterns
Input(x) : [-15 -10 -5 0 5 10 15]
Desired Output(y) : function y = 0.05 x 0.2 x 3 x + 20 for input(x)
3

[-148.7500 -20.0000 23.7500 20.0000 6.2500 20.0000 98.7500 ]

2. Architecture of MLP
W2
W3

W1

Input
Layer

First
Hidden
Layer

Second
Hidden
Layer

Output
Layer

Figure 1. Archtecture of MLP

3. Test Patterns
Input(x) : values from -15 to 15
Desired Output(y) : function y = 0.05 x 0.2 x 3 x + 20
3

Training Vectors
100
input point
desired output y=0.05*x 3-0.2*x 2-3*x+20
50
g
ni
n
ni
ar
t
r
of
y
r
ot
c
e
V
t
e
g
r
a
T

-50

-100

-150
-15

-10

-5

0
Input Vector x

10

15

Figure 2. input data


Performance is 9.99124e-007, Goal is 1e-006

10

10

k
c
al
Bl
a
o
G
e
ul
Bg
ni
ni
ar
T

10

RMS error

-2

10

-4

10

Goal : 1e-006

-6

10

100

200

300
400
752 Epochs

500

600

700

Figure 3. RMS error values at each iterations

Testing MLP Network


100
Input point
Desired output
Test output
50

0
t
u
pt
u
O

-50

-100

-150
-15

-10

-5

0
Input Vector

10

Figure 4. Test pattern simulation

Matlab Program with toolbox :


clear all

% clear all data and initializing

% process of making a data set for MLP training


%-------------------------------------------------x=[-15 -10 -5 0 5 10 15];

% input for training

y=0.05*x.^3-0.2*x.^2-3*x+20;

% output for training

%--------------------------------------------------

% Process of drawing the picture of trained MLP


%-------------------------------------------------x1=[-15:1:15];
y1=0.05*x1.^3-0.2*x1.^2-3*x1+20;
figure(1);

% making a new figure window

plot(x,y,'b*',x1,y1,'b')

% drawing the graph (x,y), (x1,y1)

title('Training Vectors');

% a title of figure

xlabel('Input Vector x');

% x label of figure

ylabel('Target Vector y for trainning');

% y label of figure

legend({'input point','desired output y=0.05*x^3-0.2*x^2-3*x+20'})


% legend of figure

15

%--------------------------------------------------

% training the MLP Neural network


%----------------------------------% using the newff toolbox function
% input range : -15 ~ 15
% number of neurons of first hidden layer : 5, Log-Sigmoid Transfer function
% number of neurons of second hidden layer : 5, Log-Sigmoid Transfer function
% ouput layer : 1 Linear Transfer function
net = newff([-15 15],[5 5 1],{'logsig' 'logsig' 'purelin'});
% define the MLP neural network architecture
% ex) ------------------------------------------------% MLP neural network with 10 neurons at single hidden layer
%

newff([-15 15],[10 1], {logsig purelin } )

% MLP neural network with three hidden layers


%

newff([-15 15],[5 5 5 1], {logsig logsig logsig purelin } )

%--------------------------------------------------

% Setting the parameters for training


%-------------------------------------------------net.trainParam.epochs = 10000;
% Maximum epochs is 10000 ( iteration number of optimization )
net.trainParam.goal = 0.000001;

% Error limit is 0.000001

%--------------------------------------------------

% training the MLP by using a training pattern


%-------------------------------------------------net = train(net,x1,y1);
% training the MLP neural network by using a train function
%--------------------------------------------------

% To draw the result making the test pattern


%-------------------------------------------------Test=[-15:0.1:15];
Output=sim(net,Test);

% making the input of test pattern


% simulate the MLP network by using a

% Output is the result of simulation

sim function

%--------------------------------------------------

% drawing the figure of result of simulation


%-------------------------------------------------figure(2);
plot(x,y,'b*',x1,y1,'b',Test,Output,'r')
title('Testing MLP Network');
xlabel('Input Vector');
ylabel('Output');
legend({'Input point','Desired output','Test output'})
%--------------------------------------------------

You might also like