DSReg: Provably Recovering Individual World Latents without Reconstruction
DSReg provably recovers individual world latents without reconstruction by regularizing dependency sparsity.
DSReg, or Dependency-Sparsity Regularization, aims to recover individual world latents up to signed permutation without reconstruction, a decoder, or labels. Building on LeJEPA's linear identifiability, the authors rely on Structural Diversity: different latents leave distinct dependency footprints on observations. They argue this condition is strictly weaker than structural assumptions in earlier identifiable latent-variable models, and that DSReg can be applied post hoc to trained checkpoints. Across synthetic regimes, world-model probes, visual encoders, and external renderers, they report preserved dense prediction and better individual-latent recovery.
- DSReg needs no reconstruction, decoder, or labels.
- Structural Diversity is claimed weaker than prior identifiability conditions.
- Method applies post hoc to linearly identified checkpoints, including LeJEPA.
- Tests cover synthetic data, world-model probes, encoders, and renderers.
Full article205 words · extracted from huggingface.co · click to collapse
Methods that recover individual latent variables of the world, from nonlinear ICA to dictionary learning and causal representation learning, anchor the latents to observations through reconstruction, auxiliary supervision, or distributional asymmetries such as non-Gaussianity. Methods without these anchors, including joint-embedding predictive architectures (JEPAs), identify the latent state only up to a linear transformation, so individual latents remain mixed. We close this gap: individual world latents can be provably recovered with no reconstruction, no decoder, and no labels. The key condition is Structural Diversity: different latents leave distinct dependency footprints on observations, just as no two snowflakes are alike. Building on the linear identifiability that LeJEPA provides, we prove that under Structural Diversity, DSReg (Dependency-Sparsity Regularization) recovers individual world latents up to signed permutation, without reconstruction or a decoder. It applies post hoc to any linearly identified representation, reusing trained checkpoints at no loss over joint training, and establishes the first fully identifiable JEPA that recovers every world latent. Moreover, as a condition on dependency footprints, Structural Diversity is strictly weaker than all structural conditions of prior identifiable latent variable models. Across synthetic regimes, world model probes, learned visual encoders, and external renderers, DSReg preserves dense prediction while improving individual-latent recovery and downstream use with scales.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2610.09457