NeurIPS 2020

The Strong Screening Rule for SLOPE

Meta Review

The work is well presented, motivated. Applications are interesting and speedups are demonstrated for computing the Slope estimator. The contribution proceeds from a reformulation of the subdifferential of the Slope penalty, and the adaptation of Tibshirani et al. "Strong rule for the Lasso" to this penalty. To improve the general reading, I recommend clarifying the point raised by R#4: "what are the advantages of the subgradients found by this work compared to that of [11]" in the camera ready version.