how the regualrized term R(s) is performed
HXYNODE opened this issue · comments
thank you for your awesome work! the issue just as the title. i cannot find the corresponding code related with the regularization term 'R(s)' for encouraging model to incorporate more pseudo-labeled data in your released code. Could you introduce the formulation of the regularization term proposed in your paper in details and help me check if i missed any key points?
@HXYNODE Thanks for your interest in our paper.
The regularization term 'R(s)' is our manually designed scheme that we select top percent of reliable data into the model training. The symbol R(s) is only used for a theoretical formulation for optimization. Otherwise there is a trivial solution that all s_i = 0.
So in the code implementation, we do not add any special code for the term symbol R(s). Indeed you may treat the increasing selection ratio as the R(s).
i got it! appreciated your reply so quickly. thanks again!