The authors developed LeHami, a joint-embedding predictive architecture whose latent state is divided into a canonical pair, q and p, analogous to position and momentum.
Symplectic latent dynamics prevent rollout contraction in visual world models
LeHami uses Hamiltonian structure to preserve latent phase-space volume while improving long-horizon motion prediction on visual pendulum tasks.
Academic
Romina Yalovetzky · Randall Balestriero · Yann LeCun
Courant Institute, New York University · Brown University · AMI Labs
Research Digest··3 min read
Yalovetzky, Balestriero and LeCun replace an unconstrained JEPA predictor with a learned, action-conditioned Hamiltonian flow over position-like and momentum-like latent variables.
Why this paper
From Courant Institute, New York University and 2 others · Released code
In one line
A symplectic Hamiltonian flow as the JEPA predictor preserves latent phase-space volume and substantially improves long-horizon velocity prediction on pendulum environments.
What it released
Code
What we could check
- ✓Code link in the paper (github.com)
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·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.
§