Zhuliang Yao comments

Results 5 comments of


                                            Zhuliang Yao

Bert Data for Pretraining: No such file or directory: 'bert_data/validation_512_only'

Hi @skaarthik, have you decided to update the [zip-dataset](https://bertonazuremlwestus2.blob.core.windows.net/public2/bert_data.tar.gz) or the data prep instruction? Besides, I wonder what if I did as @usuyama suggested? Will there be any performance influence/drop?...

What's the format of the raw data?

Thank you!

Reproduce ENAS results on RNN

@carpedm20 As far as I know, the E[Reward(m, omega)] should be calculated under the meaning of expectation, which means you are supposed to sample several models and average those rewards...

Eq(9) in paper

1. This equation is initially described at Page.5, and we put the proof at Appendix.C. 2. λ is a normalization scaler to keep the sum over Ω equals 1.

What dose the 'fixed BN' mean specifically in your paper?

'fixed BN' means that only BN is frozen (use the params from imagenet-pretrained model).