parameter

[cmn-bits]

Dear author, I apologize for the disturbance. According to the paper, the default parameter setting is α=0.2 for SST2. However, it seems that Table 1 can be approximately reproduced using α=0.1. I would like to inquire about the detailed parameter settings for the experimental results. I look forward to your response. Thank you.

[huangtiansheng]

Good question. I think we have made some refinement to the code after paper acceptance such that it wlll have some slight difference with the number in the paper (this refinement should have better performance).

For the current code, the best mask ratio of SST2 is set as alpha=0.1 and I attach the training log (for the current code) for you to verify.

antidote_poison_ratio-2615092.txt
antidote_poison_ratio-2615093.txt
antidote_poison_ratio-2615094.txt
antidote_poison_ratio-2615095.txt
antidote_poison_ratio-2615091.txt