Information Technology Reference
In-Depth Information
Fig. 15 The variation of MSE with the training epochs
the training process. MSE is an accepted measure of the performance index often
used in backpropagation MLP networks. The lower value of MSE indicates that the
network is capable of mapping the input and output accurately. The accepted error
level has been set to 0.001 and the training will stop when the
final value of MSE
reaches at 0.001 or below this level. The performance value indicates the extent of
training of the network. A low performance value indicates that the network has
been trained properly.
The number of epochs required to train a network also indicates the network
performance. The adjustable parameters of the network will not converge properly
if the number of training epochs is insuf
cient and the network will not be well
trained. On the other hand, the network will take unnecessary long training time if
there are excessive training epochs. The number of training epochs should be
suf
cient enough so as to meet the aim of training. The maximum allowed epochs
for the training process has been set to 100,000 (One Lac). If the network could not
converge within the maximum allowed epochs count, the training will stop.
The network was trained with 50 samples of handwritten characters where each
sample has 26 characters (a
×
26 = 1,300)
handwritten characters have been involved. The process of successfully training the
neural network by first training sample can be seen in Fig.
15
. It is clear from the
-
z). In the proposed experiment, 1,300 (50
figure that the training has properly converged to the goal after 184 epochs.
If there is a saturated or horizontal straight line at the end of the MSE plot, the
further training epochs are no longer bene
cial and the training should be stopped
by introducing the stopping criterion such as maximum number of training epochs
allowed in addition to the stopping criterion of maximum acceptable error as
Search WWH ::
Custom Search