The authors analyze joint-embedding predictive architectures for systems whose unstable modes amplify small disturbances unless corrected by feedback.
Inverse dynamics helps visual world models retain unstable control modes
Adding action-sequence reconstruction prevents learned representations from discarding state directions that feedback controllers need to stabilize.
Top University
Leonardo F. Toso · Yann LeCun · James Anderson · Oumayma Bounou
Columbia University · New York University · AMI Labs
Research Digest··2 min read
Toso and colleagues identify a control failure hidden by standard JEPA training: accurate next-step prediction and anti-collapse regularization can coexist with representations that erase unstable but controllable dynamics.
Why this paper
From Columbia University and 2 others
In one line
Adding an inverse dynamics loss to JEPA world models preserves unstable control modes that standard prediction objectives collapse.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§