Why using the labeled and unlabeled split?

[Xiangyi1996]

You separate the images in each class into disjoint labeled and unlabeled sets. Why we can't sample labeled images and take the remained images as unlabeled split and sample unlabeled data from here? In such way we don't need to create the labeled and unlabeled split. Since the update of prototypes among tasks do not influence others.

[renmengye]

We have a split in labeled and unlabeled because it is similar to the regular notion of semi-supervised learning. Using the rest of the query set is definitely possible, which are referred to as "transductive" setting in the literature.