Announcement
DSReg is claimed to recover individual latent variables without reconstruction
A post says the regularization can be applied after pretraining and cites JEPA as a no-reconstruction example.
TLDR
A post claims DSReg, a simple regularization that can be applied to a pretrained model after the fact, can provably recover individual latent variables of the true world even without reconstruction. It cites JEPA as an example of the latter.
Combined views
14.1K
2 Sources, first seen ago
