Learning to Mutate with Hypergradient Guided Population

Neural Information Processing Systems 

Computing the gradient of model hyperparameters, i.e., hypergradient, enables a