DSReg enables latent variable recovery without reconstruction

DSReg provides a method to provably recover individual world latents without using decoders, reconstruction, or labels. It relies on the principle of Structural Diversity, where different latents leave unique dependency footprints on observations to ensure identifiability.