Small zero-drift latent errors can regularize world models, and Jacobian regularization helps when latent errors have nonzero drift.
Day- dreamer: World models for physical robot learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Towards Unraveling and Improving Generalization in World Models
Small zero-drift latent errors can regularize world models, and Jacobian regularization helps when latent errors have nonzero drift.