Lecture 10
Lecture 10
Backpropagation Algorithm
Basic Neuron Model In A
Feedforward Network
Total-Sum-Squared-Error (TSSE)
1
TSSE (desired actual ) 2
2 patterns outputs
Root-Mean-Squared-Error (RMSE)
2 * TSSE
RMSE
# patterns*# outputs
– Bias weights
Outputs
only the identity function
Inputs
Outputs
Inputs
and
net j bias *Wbias O pkWkj
k
k ranges over the input indices
and Wjk is the weight on the
connection from input k to
neuron j
pj O pj (1 O pj ) pkWkj
k
output(s)
inputs
0.1 Target
output
Sample 1
input 0.9
0.9 1
Selectan architecture
Randomly initialize weights
While error is too large
– Select training pattern and feedforward to find actual
network output
– Calculate errors and backpropagate error signals
– Adjust weights
Evaluate performance using the test set
?? Actual Target
0.1 ?? output output
Sample ?? ??
1
input ??? 0.9
??
0.9 ?? ?? 1
??
?? 1