Statistical Benefits of Fine-Tuning from Pretrained Initialization in Diagonal Linear Networks

Published:

Direct Link