CLASSIFICATION AND PREDICTION - Universität Wien
CLASSIFICATION AND PREDICTION - Universität Wien
CLASSIFICATION AND PREDICTION - Universität Wien
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
(19)<br />
?X?<br />
to the previous layer, 2<br />
Peter Brezany Institut für Softwarewissenschaft, WS 2002 12<br />
Backpropagation Algorithm<br />
Algorithm: Backpropagation. Neural network learning for classification,<br />
using the backpropagation algorithm.<br />
Input: The training samples, samples; the learning rate, ;<br />
a multilayer feed-forward network, network.<br />
Slide 23<br />
Output: A neural network trained to classify the samples.<br />
Method:<br />
(1) Initialize all weights and biases in network;<br />
(2) while terminating condition is not satisfied<br />
(3) for each training sample in samples<br />
(4) // Propagate the inputs forward:<br />
!<br />
"$#&%('*),+ ) #.- )/10<br />
(5) for each hidden or output layer unit<br />
(6) ; //compute the net input of unit with respect<br />
-3#4% 5<br />
(7)<br />
#>=?<br />
// compute the output of each unit<br />
57698;:9<<br />
Backpropagation Algorithm (2)<br />
(8) // Backpropagate the errors:<br />
(9) for each unit in the output layer<br />
(10) @!ABA#&%(-3#§CEDGFH-I#.JKCMLN#!FO-3#J ; // compute the error<br />
(11) for each unit in the hidden layers, from the last to the first hidden layer<br />
Slide 24<br />
(12) @!ABA#&%(-3#§CEDGFH-I#.J '*P @QABA P + # P ; //compute the error with respect to<br />
the next higher layer, R<br />
+ ) # <br />
) S&+ ) #Q%TCMJE@!ABA#U-<br />
(13) for each weight in network<br />
(14) ; // weight increment<br />
(15) + ) #Q%V+ ) # / S&+ ) # ;<br />
?<br />
// weight update<br />
0 # <br />
S #&%TCMJW@QABA# 0<br />
(16) for each bias in network<br />
(17) ; // bias increment<br />
(18) 0 #4% 0 # / S 0 # ;<br />
?<br />
// bias update