Reviews: Scaling Memory-Augmented Neural Networks with Sparse Reads and Writes

The novelty in this work is thin, see e.g. On the other hand, I have not seen such tools used for writeable memories, and even if there is not a conceptual leap in this paper, certainly the practical aspects of making it work are worth reporting. Nevertheless, I do recommend accepting the paper, but would like to see some changes. Especially the descriptions of the tasks are lacking; considering the code for the tasks from the original NTM paper have not been released, and there is no "standard" version of the tasks, it is crucial that the paper give careful descriptions of the construction of the tasks. I would also ask that the authors commit to releasing the code for their experiments.