The authors developed H-JEPA, an end-to-end training method in which multiple action-conditioned predictive models learn at progressively longer timescales.
Hierarchical latent models improve long-horizon visual planning with less compute
H-JEPA learns separate predictive representations across timescales, then uses higher-level predictions as subgoals for lower-level planners.
Top University
Wancong Zhang · Basile Terver · Michael Rabbat · Yann LeCun · Randall Balestriero
NYU · Advanced Machine Intelligence · INRIA Paris · Brown University
Research Digest··3 min read
Zhang et al.
Why this paper
From INRIA Paris and 3 others
In one line
Hierarchical world models with separate latent spaces at each level improve long-horizon planning success and efficiency over flat models.
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 (2 noted)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§