9e1a36515d6704d7eb7a30d783400e5d-AuthorFeedback.pdf

Reviewers found that this work is solid, the main idea is interesting and clearly presented.2 This is a8 new design of tensor-train decomposition for spatio-temporal data. We add a new comparison between ours and9 [40] in the table below. We agree that there is a16 trade-offbetween FLOPs and latency. In our current implementation, we use Alg. 2 to reduce the GPU memory requirement.