You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
According to the paper, ReZero initializes each layer to perform identity operation.
It seems that ReZero is designed for training networks from scratch. I wonder is it applicable to fine-tuning and improve convergence?
The text was updated successfully, but these errors were encountered:
According to the paper, ReZero initializes each layer to perform identity operation.
It seems that ReZero is designed for training networks from scratch. I wonder is it applicable to fine-tuning and improve convergence?
The text was updated successfully, but these errors were encountered: