"Why is learning rate decay not commonly used in partial-data experiments in semi-supervised settings?

[1224wxwx]

Hi, thanks for sharing your amazing project. I have a question about the learing rate settings.

Generally, when training template detection models, we often use the strategy of learning rate decay. However, why is it that in a semi-supervised setting, I've observed that this is not commonly done?