The idea of GAs is very straightforward—evaluate the current generations, use only the top-performing individuals to generate the next candidate solutions, and discard the other individuals. This is shown in the preceding diagram. The survivors will generate the next population by crossover and mutation. These two processes are represented in the following diagram. Crossover is done by selecting two solutions among the survivors and combining their parameters. Mutation, on the other hand, involves changing a few random parameters on the offspring's genotype:

Crossover and mutation can be approached in many different ways. In the simpler version, crossover is done by choosing parts from the two parents randomly, and mutation is done by mutating the solution that's obtained by adding Gaussian noise with a fixed standard deviation. By only keeping the best individuals and injecting their genes into the newly born individuals, the solutions will improve over time until a condition is met. However, on complex problems, this simple solution is prone to be stuck in a local optimum (meaning that the solution is only within a small set of candidate solutions). In this case, a more advanced genetic algorithm such as NeroEvolution of Augmenting Topologies (NEAT) is preferred. NEAT not only alters the weights of the network but also its structure.