Export Reviews, Discussions, Author Feedback and Meta-Reviews

The authors uses an estimation approach to estimate the scaling matrix and replaces an O(np p 3) per iteration cost with an O(np p 2) cost. The new optimization method applies to generalized linear models and is extremely well crafted. The paper is extremely clear, the quality is superb, the originality is good and the significance only limited by the experiments and the importance of generalized linear models. The experiments are a bit weak. There are only two examples -- logistic regression and ordinary least squares.