After training the network, we can use it (recalling phase) in a recognition task...

Recognition (input pattern without noise) :

Recognition (input patern with noise) :

Real-world applications cannot avoid noise. However, neural networks recognize noisy input patterns thanks to generalization of the training patterns. Next figure shows a noisy six clearly recognized.

Too much noise... :

Neural networks are robust in the presence of noise, but there is always a limit...
This example shows a highly noisy pattern, which can be considered either as a corrupted 5 or 6;
the network is unable to correctly classify the input, however, neither can a human observer.

END of this tutorial.....

Version 16.I.96