I found the scale_grad_by_freq option in the nn.Embedding module.
However, I couldn’t find the reference for that option.
Does anyone know which survey or paper used this technique?
Thank you.
I found the scale_grad_by_freq option in the nn.Embedding module.
However, I couldn’t find the reference for that option.
Does anyone know which survey or paper used this technique?
Thank you.
Can anyone provide some updates on this? I wanted to know the motivations and the results of using this technique for learning embeddings…