L2 to L1 training scheme

[jdsjfi]

Hello, the L2 to L1 training scheme presented in your paper is interesting, but I have some questions. In the paper, it says, "p is linearly reduced from 2 to 1", what does it mean? Does it mean that p is linearly reduced from 2 to 1 with respect to each epoch or each step? Thanks very much!!!

[HantingChen]

Yes

[jdsjfi]

Sorry, I think you may have misunderstood what I mean. I reproduce L2 to L1 training scheme by reducing p for each step. I want to know whether my apprach is correct. Thank you very much!

[HantingChen]

p is reduced for each epoch, but I think it also works if you reduce it for each step.