@yuranusduke
Great work! I have tried my own implementation -Please check it here- but my model keeps getting NaNs everywhere. Luckily I found yours. I am asking if you have tried to train the model and get results? Just to make sure that the paper has given us enough information to implement it. I looked at the differences between my code and yours and there is not much of a change or that I have made a huge mistake. I am not an expert, I am just a fresh graduate working on improving my skills.
I tried to contact you through e-mail address but gmail doesn't send email to 163.com