Causal-JEPA

Causal-JEPA: Learning World Models through Object-Level Latent Masking

Heejeong Nam, Quentin Le Lidec, Lucas Maes, Yann LeCun, Randall Balestriero · arXiv 2026

Read paper · arXiv:2602.11389

Mask object-level latent trajectories to infer their dynamics from other objects.

What you will implement

Object masking with initial identity anchors and history/future prediction losses.

• 60. Causal-JEPA Object Trajectory Masking

  Mask object trajectories across time, preserve first-frame identity anchors, and implement masked-history plus future prediction against frozen targets.

  Object slots, projection weights and time embeddings are supplied; implement input assembly and losses without training an image encoder or full transformer.