Where is the back propagation? and is its target?

[olixis]

I could not find where you are training the neural network, i see the activation and the GA stuff but not training.

[Ttommeke]

He doesn't. The approach in this project is to generate a bunch of random networks, put them to the test and grade their performance.

Then the best of these random networks are kept and the values of their biases are mixed to create (hopefully) a beter network. To add a little bit of variety and help the "spicies" evolve, the biases and connections are changed a little bit by a random factor. This process is repeated over and over again.

The idea behind this approach is that you don't need training data to let the neural net learn. Only an environment.