Neural evolution combines neural networks and evolutionary computation to construct a neural network using evolution to determine parameters, like weight values, activation functions, and loss functions, instead of backpropagation or stochastic gradient descent.