Announcement
JEPA-TTT proposes persistent test-time training for latent world models
The announcement says pretrained models’ predictions can become unreliable when test-time dynamics differ from training.
TLDR
A post introducing JEPA-TTT says it enables persistent test-time training of latent world models for planning under dynamics shifts. It says pretrained models’ predictions can become unreliable when test-time dynamics differ from training, and links to a project page and paper.
Combined views
13.1K
2 Sources, first seen ago
