Basically:

• the learning signal is made of 10 different 5x5 pictures representing a growing and shrinking dot. (the first one is `pure blank')
• the structure of the network is made so we give three pictures to the network and we want the `next' picture on the output.
• then we give the initial input values to the network, and we generate steps independently: we use the output values, for the next third of the input. (ok, this is hard to explain -- you can also have a look in the corresponding method of the main class)

In the left part of the window we provide as initial values, the values corresponding to the learning signal, so we expect a nice and smooth signal.

In the right part of the window we provide as initial values, wrong values (the dot is growing too fast), and we want the network to regain automatically the original signal.