Question about the 1-D case

[skyshine102]

Hi, I found your paper very interesting and I have a quick question. Here, the paper proposed CPE for the vision task, which is 2D and therefore the locality assumption would be valid. I would like to ask whether u have designed 1-D CPE and do similar experiments in the NLP tasks? If so, how could we choose the kernel size?

[cxxgtxy]

Thanks for your attention. Sorry for the late reply. Our goal is to better handle the positional encoding in vision transformer. We don't do any experiments on NLP tasks. We believe it's an interesting topic.