Ha, David. “Neural Network Evolution Playground with Backprop NEAT.” (2016).
URL1 URL2
Most students studying machine learning learn that in order to train a neural network, one should define an objective function to measure how well the neural network is performing some task, and to use back propagation to solve for the derivatives of this objective function with respective to each weight, and afterwards use these gradients to iteratively solve for a good set of weights for the neural network. This framework is known as end-to-end training.