Yann LeCun explains that I-JEPA involves masking parts of an image and training the system to predict the representation of the uncorrupted image.
in the case of I-JEPA, you don't need to do any of this, you just mask some parts of it, right? You just basically remove some regions like a big block, essentially.