The Neural Turing Machine is a neural network architecture inspired by biological working memory and digital computers that is differentiable end-to-end and can be trained with gradient descent. Experiments showed it is capable of learning simple algorithms from example data and generalizing well outside its training data.