Experiment NeuroEvolution of Augmenting Topologies (NEAT)
NEAT algorithm is a way to compute neural network by changing its topology using genetic evolution process.
Whereas traditional neural network has a fixed topology and adjust its neuron weights to approximate a solution, NEAT algorithm changes the weight but also the topology itself.
As a first test, we approximate the XOR function. It takes about 20 generations of 150 neural network to do a XOR function
Next step will be to approximate complex behavior such as human playing Mario.
Original article : http://nn.cs.utexas.edu/downloads/papers/stanley.ec02.pdf
Neat MarIO : https://www.programmersought.com/article/11446589053/