This is the arxiv link for the Projected Functional Regularization (PFR) paper:
I see the term projector is borrowed from the Barlow Twins paper. It seems like a projector is a truncated decoder? Maybe there is more nuance, but this is what I gathered from a shallow read.
So, I guess I understand what a “view projector” is and the paper mentions its loss function. But, its unclear what a “temporal projector” is. It also mentions that it is a “learned temporal projection,” so is this projector pre-trained? Does that even make sense as a concept? If so, I wasn’t able to understand how it is trained?