Open danielhers opened 7 years ago
Ma et al. (2017) introduce dropout with regularization (DR), the benefit of which in NLP is shown by Strubell et al. (2017). Does DyNet dropout support this regularization?
I doubt this is implemented, but if it is not easy to implement with the operations already provided by DyNet it might be nice to have.
Ma et al. (2017) introduce dropout with regularization (DR), the benefit of which in NLP is shown by Strubell et al. (2017). Does DyNet dropout support this regularization?