In this paper, authors propose an instance-wise feature grouping, which is an interesting research topic. The proposed algorithm is backed up with theoretical analysis, and authors compared the proposed algorithm with several existing methods. However, there exist some concerns about experiments and did not fully addressed. I strongly encourage authors to address the following issues in the final version. 1. Validating the approach on gene expression (comments from Reviewer#2) The feature grouping per instance provided something novel for gene expression. (or some other dataset) 2. Comparison to attention-based learning methods. Therefore, I recommend weak acceptance for this paper. If this paper is not accepted, it is simply due to the lack of slots for this time.