Name of Reviewer ------------------ Shulin Yang Key Contribution ------------------ This paper handled the diverse classification problem by learning the optimal trade-off between invariance and discriminative power. Novelty -------- The paper worked on a classical problem that has been studied for long. The method proposed here is based on the former theory. But yes, it is new. Reference to prior work ----------------------- The reference has included enough, I think. Clarity ------- The paper is written in a clear way, and it has illustrated clearly its relationship to previous and other related works. To fully understand the paper, I think some extent of background knowledge of statistical learning theory is needed. It's hard to read without any background in this area, and it will take too much space if all theories used here are presented in this paper. Technical Correctness --------------------- You should be able to follow each derivation in most papers. If there are certain steps which make overly large leaps, be specific here about which ones you had to skip. Experimental Validation ----------------------- There are famous three image datasets used here for experiment, which is very persuasive. In my opinion, classification on image depends more on feature extraction than on the classifier. So I am not sure whether it is very meaningful to prove its effectiveness on these image sets. Overall Evaluation ------------------ Overall, the paper is very concrete and well ordered, with sufficient experiments to support the idea. Questions and Issues for Discussion ----------------------------------- We could discuss about the effectiveness of this method in comparison to other related schemes, and the situation/application with which each method works better or would fail. We can also think about whether some existing method for classification may actually be equivalent to this one.