Is a channel sampling mask fixed?

[nabk89]

Hello, thank you for your code sharing.
When looking into your code, I have a question about implementation for your partial channel connection idea.

In your code (model_search.py), it seems that "channel_shuffle" function only choose the first quartile of channels (including "forward" function of MixedOp class).
Does it mean that a channel sampling mask S_i,j defined in your paper is a fixed mask?

Please answer my question.
Thank you!

[yuhuixu1993]

Hi, we offer two kinds of implementations in this repo. One is the mentioned channel_shuffle (it induces fixed mask each layer while different within layers.) which is an approximation; The other is a totally random mask, you can check model_search_random.py, if you are interested.

[nabk89]

Thank you, but I don't understand that working of channel_shuffle is different within layers.
Would you explain this part in a detail?

[sjjdd]

请问固定和随机哪种效果更好一些啊?