Reviews: Learning Sparse Distributions using Iterative Hard Thresholding

Post-rebuttal: I have downgraded my overall score to 7. I am troubled by the lack of motivation (and that in the rebuttal, the authors defer more discussion of model compression to future work). Also, I'd have liked to see in the rebuttal more details about the "more comprehensive discussion" regarding alternate algorithms. Apparently, this is the first work that studies this problem for general loss functions. The goal of the work is to adapt the well-known IHT algorithm, a form of projected gradient descent, to this problem. Again, as far as I can see, this approach is original to this work.