The authors begin with representations such as those produced by LeJEPA, which can contain the complete latent state while leaving its components mixed by an unknown linear transformation.
Dependency sparsity can isolate world latents without reconstructing observations
DSReg rotates linearly identified representations into individual latent factors when those factors have distinct dependency patterns.
Research Lab
Yujia Zheng · David Klindt · Randall Balestriero · Bernhard Schölkopf
University of Illinois Urbana-Champaign · Cold Spring Harbor Laboratory · Brown University · Max Planck Institute for Intelligent Systems · ELLIS Institute Tübingen
Research Digest··3 min read
Zheng et al.
Why this paper
From Max Planck Institute for Intelligent Systems and 4 others
In one line
Under Structural Diversity, DSReg recovers individual world latents up to signed permutation without reconstruction, decoders, labels, or non-Gaussian assumptions.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§