[R][1610.09027] Scaling Memory-Augmented Neural Networks with Sparse Reads and Writes [DeepMind] • /r/MachineLearning