Seed value for the datasets

[nandhiniramanan5]

What are the seed values used as input in your experiments to the code experiment.py for the data sets mist, letter, usps, ijcnn, covtype?

[anitan0925]

We conducted experiments with five seeds (1, 11, 111, 1111, 11111). However, I am not sure about the reproducibility due to the difference of GPU-architecture.
In addition to this, the latest code on github differs from the version in ICML2018 paper. Hence, it is better to revert the version to those on May 25 2018 or Feb 27 2018 if comparing a method with our method in ICML2018 paper.