the fact that https://arxiv.org/abs/1712.09913 doesn't cite https://arxiv.org/abs/1711.05101 is a little worrying, they might have to redo their adam tests in a v2